examples/l3fwd-power/main.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2018 Intel Corporation
   3  */
   4
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <stdint.h>
   8 #include <inttypes.h>
   9 #include <sys/types.h>
  10 #include <string.h>
  11 #include <sys/queue.h>
  12 #include <stdarg.h>
  13 #include <errno.h>
  14 #include <getopt.h>
  15 #include <unistd.h>
  16 #include <signal.h>
  17 #include <math.h>
  18
  19 #include <rte_common.h>
  20 #include <rte_byteorder.h>
  21 #include <rte_log.h>
  22 #include <rte_malloc.h>
  23 #include <rte_memory.h>
  24 #include <rte_memcpy.h>
  25 #include <rte_eal.h>
  26 #include <rte_launch.h>
  27 #include <rte_cycles.h>
  28 #include <rte_prefetch.h>
  29 #include <rte_lcore.h>
  30 #include <rte_per_lcore.h>
  31 #include <rte_branch_prediction.h>
  32 #include <rte_interrupts.h>
  33 #include <rte_random.h>
  34 #include <rte_debug.h>
  35 #include <rte_ether.h>
  36 #include <rte_ethdev.h>
  37 #include <rte_mempool.h>
  38 #include <rte_mbuf.h>
  39 #include <rte_ip.h>
  40 #include <rte_tcp.h>
  41 #include <rte_udp.h>
  42 #include <rte_string_fns.h>
  43 #include <rte_timer.h>
  44 #include <rte_power.h>
  45 #include <rte_spinlock.h>
  46 #include <rte_power_empty_poll.h>
  47 #include <rte_metrics.h>
  48 #include <rte_telemetry.h>
  49 #include <rte_power_pmd_mgmt.h>
  50
  51 #include "perf_core.h"
  52 #include "main.h"
  53
  54 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
  55
  56 #define MAX_PKT_BURST 32
  57
  58 #define MIN_ZERO_POLL_COUNT 10
  59
  60 /* 100 ms interval */
  61 #define TIMER_NUMBER_PER_SECOND           10
  62 /* (10ms) */
  63 #define INTERVALS_PER_SECOND             100
  64 /* 100000 us */
  65 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
  66 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
  67
  68 #define APP_LOOKUP_EXACT_MATCH          0
  69 #define APP_LOOKUP_LPM                  1
  70 #define DO_RFC_1812_CHECKS
  71
  72 #ifndef APP_LOOKUP_METHOD
  73 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  74 #endif
  75
  76 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  77 #include <rte_hash.h>
  78 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  79 #include <rte_lpm.h>
  80 #else
  81 #error "APP_LOOKUP_METHOD set to incorrect value"
  82 #endif
  83
  84 #ifndef IPv6_BYTES
  85 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
  86                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
  87 #define IPv6_BYTES(addr) \
  88         addr[0],  addr[1], addr[2],  addr[3], \
  89         addr[4],  addr[5], addr[6],  addr[7], \
  90         addr[8],  addr[9], addr[10], addr[11],\
  91         addr[12], addr[13],addr[14], addr[15]
  92 #endif
  93
  94 #define MAX_JUMBO_PKT_LEN  9600
  95
  96 #define IPV6_ADDR_LEN 16
  97
  98 #define MEMPOOL_CACHE_SIZE 256
  99
 100 /*
 101  * This expression is used to calculate the number of mbufs needed depending on
 102  * user input, taking into account memory for rx and tx hardware rings, cache
 103  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 104  * NB_MBUF never goes below a minimum value of 8192.
 105  */
 106
 107 #define NB_MBUF RTE_MAX ( \
 108         (nb_ports*nb_rx_queue*nb_rxd + \
 109         nb_ports*nb_lcores*MAX_PKT_BURST + \
 110         nb_ports*n_tx_queue*nb_txd + \
 111         nb_lcores*MEMPOOL_CACHE_SIZE), \
 112         (unsigned)8192)
 113
 114 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 115
 116 #define NB_SOCKETS 8
 117
 118 /* Configure how many packets ahead to prefetch, when reading packets */
 119 #define PREFETCH_OFFSET 3
 120
 121 /*
 122  * Configurable number of RX/TX ring descriptors
 123  */
 124 #define RTE_TEST_RX_DESC_DEFAULT 1024
 125 #define RTE_TEST_TX_DESC_DEFAULT 1024
 126
 127 /*
 128  * These two thresholds were decided on by running the training algorithm on
 129  * a 2.5GHz Xeon. These defaults can be overridden by supplying non-zero values
 130  * for the med_threshold and high_threshold parameters on the command line.
 131  */
 132 #define EMPTY_POLL_MED_THRESHOLD 350000UL
 133 #define EMPTY_POLL_HGH_THRESHOLD 580000UL
 134
 135 #define NUM_TELSTATS RTE_DIM(telstats_strings)
 136
 137 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 138 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 139
 140 /* ethernet addresses of ports */
 141 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 142
 143 /* ethernet addresses of ports */
 144 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
 145
 146 /* mask of enabled ports */
 147 static uint32_t enabled_port_mask = 0;
 148 /* Ports set in promiscuous mode off by default. */
 149 static int promiscuous_on = 0;
 150 /* NUMA is enabled by default. */
 151 static int numa_on = 1;
 152 static bool empty_poll_stop;
 153 static bool empty_poll_train;
 154 volatile bool quit_signal;
 155 static struct  ep_params *ep_params;
 156 static struct  ep_policy policy;
 157 static long  ep_med_edpi, ep_hgh_edpi;
 158 /* timer to update telemetry every 500ms */
 159 static struct rte_timer telemetry_timer;
 160
 161 /* stats index returned by metrics lib */
 162 int telstats_index;
 163
 164 struct telstats_name {
 165         char name[RTE_ETH_XSTATS_NAME_SIZE];
 166 };
 167
 168 /* telemetry stats to be reported */
 169 const struct telstats_name telstats_strings[] = {
 170         {"empty_poll"},
 171         {"full_poll"},
 172         {"busy_percent"}
 173 };
 174
 175 /* core busyness in percentage */
 176 enum busy_rate {
 177         ZERO = 0,
 178         PARTIAL = 50,
 179         FULL = 100
 180 };
 181
 182 /* reference poll count to measure core busyness */
 183 #define DEFAULT_COUNT 10000
 184 /*
 185  * reference CYCLES to be used to
 186  * measure core busyness based on poll count
 187  */
 188 #define MIN_CYCLES  1500000ULL
 189 #define MAX_CYCLES 22000000ULL
 190
 191 /* (500ms) */
 192 #define TELEMETRY_INTERVALS_PER_SEC 2
 193
 194 static int parse_ptype; /**< Parse packet type using rx callback, and */
 195                         /**< disabled by default */
 196
 197 enum appmode {
 198         APP_MODE_DEFAULT = 0,
 199         APP_MODE_LEGACY,
 200         APP_MODE_EMPTY_POLL,
 201         APP_MODE_TELEMETRY,
 202         APP_MODE_INTERRUPT,
 203         APP_MODE_PMD_MGMT
 204 };
 205
 206 enum appmode app_mode;
 207
 208 static enum rte_power_pmd_mgmt_type pmgmt_type;
 209 bool baseline_enabled;
 210
 211 enum freq_scale_hint_t
 212 {
 213         FREQ_LOWER    =      -1,
 214         FREQ_CURRENT  =       0,
 215         FREQ_HIGHER   =       1,
 216         FREQ_HIGHEST  =       2
 217 };
 218
 219 struct lcore_rx_queue {
 220         uint16_t port_id;
 221         uint8_t queue_id;
 222         enum freq_scale_hint_t freq_up_hint;
 223         uint32_t zero_rx_packet_count;
 224         uint32_t idle_hint;
 225 } __rte_cache_aligned;
 226
 227 #define MAX_RX_QUEUE_PER_LCORE 16
 228 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 229 #define MAX_RX_QUEUE_PER_PORT 128
 230
 231 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
 232
 233
 234 struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 235 static struct lcore_params lcore_params_array_default[] = {
 236         {0, 0, 2},
 237         {0, 1, 2},
 238         {0, 2, 2},
 239         {1, 0, 2},
 240         {1, 1, 2},
 241         {1, 2, 2},
 242         {2, 0, 2},
 243         {3, 0, 3},
 244         {3, 1, 3},
 245 };
 246
 247 struct lcore_params *lcore_params = lcore_params_array_default;
 248 uint16_t nb_lcore_params = RTE_DIM(lcore_params_array_default);
 249
 250 static struct rte_eth_conf port_conf = {
 251         .rxmode = {
 252                 .mq_mode        = RTE_ETH_MQ_RX_RSS,
 253                 .split_hdr_size = 0,
 254                 .offloads = RTE_ETH_RX_OFFLOAD_CHECKSUM,
 255         },
 256         .rx_adv_conf = {
 257                 .rss_conf = {
 258                         .rss_key = NULL,
 259                         .rss_hf = RTE_ETH_RSS_UDP,
 260                 },
 261         },
 262         .txmode = {
 263                 .mq_mode = RTE_ETH_MQ_TX_NONE,
 264         }
 265 };
 266
 267 static uint32_t max_pkt_len;
 268
 269 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 270
 271
 272 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 273
 274 #ifdef RTE_ARCH_X86
 275 #include <rte_hash_crc.h>
 276 #define DEFAULT_HASH_FUNC       rte_hash_crc
 277 #else
 278 #include <rte_jhash.h>
 279 #define DEFAULT_HASH_FUNC       rte_jhash
 280 #endif
 281
 282 struct ipv4_5tuple {
 283         uint32_t ip_dst;
 284         uint32_t ip_src;
 285         uint16_t port_dst;
 286         uint16_t port_src;
 287         uint8_t  proto;
 288 } __rte_packed;
 289
 290 struct ipv6_5tuple {
 291         uint8_t  ip_dst[IPV6_ADDR_LEN];
 292         uint8_t  ip_src[IPV6_ADDR_LEN];
 293         uint16_t port_dst;
 294         uint16_t port_src;
 295         uint8_t  proto;
 296 } __rte_packed;
 297
 298 struct ipv4_l3fwd_route {
 299         struct ipv4_5tuple key;
 300         uint8_t if_out;
 301 };
 302
 303 struct ipv6_l3fwd_route {
 304         struct ipv6_5tuple key;
 305         uint8_t if_out;
 306 };
 307
 308 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 309         {{RTE_IPV4(100,10,0,1), RTE_IPV4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 310         {{RTE_IPV4(100,20,0,2), RTE_IPV4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 311         {{RTE_IPV4(100,30,0,3), RTE_IPV4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 312         {{RTE_IPV4(100,40,0,4), RTE_IPV4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 313 };
 314
 315 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 316         {
 317                 {
 318                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 319                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 320                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 321                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 322                          1, 10, IPPROTO_UDP
 323                 }, 4
 324         },
 325 };
 326
 327 typedef struct rte_hash lookup_struct_t;
 328 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 329 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 330
 331 #define L3FWD_HASH_ENTRIES      1024
 332
 333 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 334 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 335 #endif
 336
 337 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 338 struct ipv4_l3fwd_route {
 339         uint32_t ip;
 340         uint8_t  depth;
 341         uint8_t  if_out;
 342 };
 343
 344 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 345         {RTE_IPV4(1,1,1,0), 24, 0},
 346         {RTE_IPV4(2,1,1,0), 24, 1},
 347         {RTE_IPV4(3,1,1,0), 24, 2},
 348         {RTE_IPV4(4,1,1,0), 24, 3},
 349         {RTE_IPV4(5,1,1,0), 24, 4},
 350         {RTE_IPV4(6,1,1,0), 24, 5},
 351         {RTE_IPV4(7,1,1,0), 24, 6},
 352         {RTE_IPV4(8,1,1,0), 24, 7},
 353 };
 354
 355 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 356
 357 typedef struct rte_lpm lookup_struct_t;
 358 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 359 #endif
 360
 361 struct lcore_conf {
 362         uint16_t n_rx_queue;
 363         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 364         uint16_t n_tx_port;
 365         uint16_t tx_port_id[RTE_MAX_ETHPORTS];
 366         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 367         struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
 368         lookup_struct_t * ipv4_lookup_struct;
 369         lookup_struct_t * ipv6_lookup_struct;
 370 } __rte_cache_aligned;
 371
 372 struct lcore_stats {
 373         /* total sleep time in ms since last frequency scaling down */
 374         uint32_t sleep_time;
 375         /* number of long sleep recently */
 376         uint32_t nb_long_sleep;
 377         /* freq. scaling up trend */
 378         uint32_t trend;
 379         /* total packet processed recently */
 380         uint64_t nb_rx_processed;
 381         /* total iterations looped recently */
 382         uint64_t nb_iteration_looped;
 383         /*
 384          * Represents empty and non empty polls
 385          * of rte_eth_rx_burst();
 386          * ep_nep[0] holds non empty polls
 387          * i.e. 0 < nb_rx <= MAX_BURST
 388          * ep_nep[1] holds empty polls.
 389          * i.e. nb_rx == 0
 390          */
 391         uint64_t ep_nep[2];
 392         /*
 393          * Represents full and empty+partial
 394          * polls of rte_eth_rx_burst();
 395          * ep_nep[0] holds empty+partial polls.
 396          * i.e. 0 <= nb_rx < MAX_BURST
 397          * ep_nep[1] holds full polls
 398          * i.e. nb_rx == MAX_BURST
 399          */
 400         uint64_t fp_nfp[2];
 401         enum busy_rate br;
 402         rte_spinlock_t telemetry_lock;
 403 } __rte_cache_aligned;
 404
 405 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
 406 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
 407 static struct rte_timer power_timers[RTE_MAX_LCORE];
 408
 409 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
 410 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
 411                 unsigned int lcore_id, uint16_t port_id, uint16_t queue_id);
 412
 413
 414 /*
 415  * These defaults are using the max frequency index (1), a medium index (9)
 416  * and a typical low frequency index (14). These can be adjusted to use
 417  * different indexes using the relevant command line parameters.
 418  */
 419 static uint8_t  freq_tlb[] = {14, 9, 1};
 420
 421 static int is_done(void)
 422 {
 423         return quit_signal;
 424 }
 425
 426 /* exit signal handler */
 427 static void
 428 signal_exit_now(int sigtype)
 429 {
 430
 431         if (sigtype == SIGINT)
 432                 quit_signal = true;
 433
 434 }
 435
 436 /*  Freqency scale down timer callback */
 437 static void
 438 power_timer_cb(__rte_unused struct rte_timer *tim,
 439                           __rte_unused void *arg)
 440 {
 441         uint64_t hz;
 442         float sleep_time_ratio;
 443         unsigned lcore_id = rte_lcore_id();
 444
 445         /* accumulate total execution time in us when callback is invoked */
 446         sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
 447                                         (float)SCALING_PERIOD;
 448         /**
 449          * check whether need to scale down frequency a step if it sleep a lot.
 450          */
 451         if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
 452                 if (rte_power_freq_down)
 453                         rte_power_freq_down(lcore_id);
 454         }
 455         else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
 456                 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
 457                 /**
 458                  * scale down a step if average packet per iteration less
 459                  * than expectation.
 460                  */
 461                 if (rte_power_freq_down)
 462                         rte_power_freq_down(lcore_id);
 463         }
 464
 465         /**
 466          * initialize another timer according to current frequency to ensure
 467          * timer interval is relatively fixed.
 468          */
 469         hz = rte_get_timer_hz();
 470         rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
 471                                 SINGLE, lcore_id, power_timer_cb, NULL);
 472
 473         stats[lcore_id].nb_rx_processed = 0;
 474         stats[lcore_id].nb_iteration_looped = 0;
 475
 476         stats[lcore_id].sleep_time = 0;
 477 }
 478
 479 /* Enqueue a single packet, and send burst if queue is filled */
 480 static inline int
 481 send_single_packet(struct rte_mbuf *m, uint16_t port)
 482 {
 483         uint32_t lcore_id;
 484         struct lcore_conf *qconf;
 485
 486         lcore_id = rte_lcore_id();
 487         qconf = &lcore_conf[lcore_id];
 488
 489         rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
 490                         qconf->tx_buffer[port], m);
 491
 492         return 0;
 493 }
 494
 495 #ifdef DO_RFC_1812_CHECKS
 496 static inline int
 497 is_valid_ipv4_pkt(struct rte_ipv4_hdr *pkt, uint32_t link_len)
 498 {
 499         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 500         /*
 501          * 1. The packet length reported by the Link Layer must be large
 502          * enough to hold the minimum length legal IP datagram (20 bytes).
 503          */
 504         if (link_len < sizeof(struct rte_ipv4_hdr))
 505                 return -1;
 506
 507         /* 2. The IP checksum must be correct. */
 508         /* if this is not checked in H/W, check it. */
 509         if ((port_conf.rxmode.offloads & RTE_ETH_RX_OFFLOAD_IPV4_CKSUM) == 0) {
 510                 uint16_t actual_cksum, expected_cksum;
 511                 actual_cksum = pkt->hdr_checksum;
 512                 pkt->hdr_checksum = 0;
 513                 expected_cksum = rte_ipv4_cksum(pkt);
 514                 if (actual_cksum != expected_cksum)
 515                         return -2;
 516         }
 517
 518         /*
 519          * 3. The IP version number must be 4. If the version number is not 4
 520          * then the packet may be another version of IP, such as IPng or
 521          * ST-II.
 522          */
 523         if (((pkt->version_ihl) >> 4) != 4)
 524                 return -3;
 525         /*
 526          * 4. The IP header length field must be large enough to hold the
 527          * minimum length legal IP datagram (20 bytes = 5 words).
 528          */
 529         if ((pkt->version_ihl & 0xf) < 5)
 530                 return -4;
 531
 532         /*
 533          * 5. The IP total length field must be large enough to hold the IP
 534          * datagram header, whose length is specified in the IP header length
 535          * field.
 536          */
 537         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct rte_ipv4_hdr))
 538                 return -5;
 539
 540         return 0;
 541 }
 542 #endif
 543
 544 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 545 static void
 546 print_ipv4_key(struct ipv4_5tuple key)
 547 {
 548         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
 549                 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
 550                                 key.port_dst, key.port_src, key.proto);
 551 }
 552 static void
 553 print_ipv6_key(struct ipv6_5tuple key)
 554 {
 555         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 556                 "port dst = %d, port src = %d, proto = %d\n",
 557                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 558                 key.port_dst, key.port_src, key.proto);
 559 }
 560
 561 static inline uint16_t
 562 get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
 563                 lookup_struct_t * ipv4_l3fwd_lookup_struct)
 564 {
 565         struct ipv4_5tuple key;
 566         struct rte_tcp_hdr *tcp;
 567         struct rte_udp_hdr *udp;
 568         int ret = 0;
 569
 570         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 571         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 572         key.proto = ipv4_hdr->next_proto_id;
 573
 574         switch (ipv4_hdr->next_proto_id) {
 575         case IPPROTO_TCP:
 576                 tcp = (struct rte_tcp_hdr *)((unsigned char *)ipv4_hdr +
 577                                         sizeof(struct rte_ipv4_hdr));
 578                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 579                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 580                 break;
 581
 582         case IPPROTO_UDP:
 583                 udp = (struct rte_udp_hdr *)((unsigned char *)ipv4_hdr +
 584                                         sizeof(struct rte_ipv4_hdr));
 585                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 586                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 587                 break;
 588
 589         default:
 590                 key.port_dst = 0;
 591                 key.port_src = 0;
 592                 break;
 593         }
 594
 595         /* Find destination port */
 596         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 597         return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
 598 }
 599
 600 static inline uint16_t
 601 get_ipv6_dst_port(struct rte_ipv6_hdr *ipv6_hdr, uint16_t portid,
 602                         lookup_struct_t *ipv6_l3fwd_lookup_struct)
 603 {
 604         struct ipv6_5tuple key;
 605         struct rte_tcp_hdr *tcp;
 606         struct rte_udp_hdr *udp;
 607         int ret = 0;
 608
 609         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 610         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 611
 612         key.proto = ipv6_hdr->proto;
 613
 614         switch (ipv6_hdr->proto) {
 615         case IPPROTO_TCP:
 616                 tcp = (struct rte_tcp_hdr *)((unsigned char *) ipv6_hdr +
 617                                         sizeof(struct rte_ipv6_hdr));
 618                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 619                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 620                 break;
 621
 622         case IPPROTO_UDP:
 623                 udp = (struct rte_udp_hdr *)((unsigned char *) ipv6_hdr +
 624                                         sizeof(struct rte_ipv6_hdr));
 625                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 626                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 627                 break;
 628
 629         default:
 630                 key.port_dst = 0;
 631                 key.port_src = 0;
 632                 break;
 633         }
 634
 635         /* Find destination port */
 636         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 637         return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
 638 }
 639 #endif
 640
 641 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 642 static inline uint16_t
 643 get_ipv4_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
 644                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 645 {
 646         uint32_t next_hop;
 647
 648         return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 649                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 650                         next_hop : portid);
 651 }
 652 #endif
 653
 654 static inline void
 655 parse_ptype_one(struct rte_mbuf *m)
 656 {
 657         struct rte_ether_hdr *eth_hdr;
 658         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 659         uint16_t ether_type;
 660
 661         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 662         ether_type = eth_hdr->ether_type;
 663         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
 664                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 665         else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6))
 666                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 667
 668         m->packet_type = packet_type;
 669 }
 670
 671 static uint16_t
 672 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 673                struct rte_mbuf *pkts[], uint16_t nb_pkts,
 674                uint16_t max_pkts __rte_unused,
 675                void *user_param __rte_unused)
 676 {
 677         unsigned int i;
 678
 679         for (i = 0; i < nb_pkts; ++i)
 680                 parse_ptype_one(pkts[i]);
 681
 682         return nb_pkts;
 683 }
 684
 685 static int
 686 add_cb_parse_ptype(uint16_t portid, uint16_t queueid)
 687 {
 688         printf("Port %d: softly parse packet type info\n", portid);
 689         if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
 690                 return 0;
 691
 692         printf("Failed to add rx callback: port=%d\n", portid);
 693         return -1;
 694 }
 695
 696 static inline void
 697 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
 698                                 struct lcore_conf *qconf)
 699 {
 700         struct rte_ether_hdr *eth_hdr;
 701         struct rte_ipv4_hdr *ipv4_hdr;
 702         void *d_addr_bytes;
 703         uint16_t dst_port;
 704
 705         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 706
 707         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 708                 /* Handle IPv4 headers.*/
 709                 ipv4_hdr =
 710                         rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
 711                                                 sizeof(struct rte_ether_hdr));
 712
 713 #ifdef DO_RFC_1812_CHECKS
 714                 /* Check to make sure the packet is valid (RFC1812) */
 715                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 716                         rte_pktmbuf_free(m);
 717                         return;
 718                 }
 719 #endif
 720
 721                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 722                                         qconf->ipv4_lookup_struct);
 723                 if (dst_port >= RTE_MAX_ETHPORTS ||
 724                                 (enabled_port_mask & 1 << dst_port) == 0)
 725                         dst_port = portid;
 726
 727                 /* 02:00:00:00:00:xx */
 728                 d_addr_bytes = &eth_hdr->dst_addr.addr_bytes[0];
 729                 *((uint64_t *)d_addr_bytes) =
 730                         0x000000000002 + ((uint64_t)dst_port << 40);
 731
 732 #ifdef DO_RFC_1812_CHECKS
 733                 /* Update time to live and header checksum */
 734                 --(ipv4_hdr->time_to_live);
 735                 ++(ipv4_hdr->hdr_checksum);
 736 #endif
 737
 738                 /* src addr */
 739                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
 740                                 &eth_hdr->src_addr);
 741
 742                 send_single_packet(m, dst_port);
 743         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 744                 /* Handle IPv6 headers.*/
 745 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 746                 struct rte_ipv6_hdr *ipv6_hdr;
 747
 748                 ipv6_hdr =
 749                         rte_pktmbuf_mtod_offset(m, struct rte_ipv6_hdr *,
 750                                                 sizeof(struct rte_ether_hdr));
 751
 752                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 753                                         qconf->ipv6_lookup_struct);
 754
 755                 if (dst_port >= RTE_MAX_ETHPORTS ||
 756                                 (enabled_port_mask & 1 << dst_port) == 0)
 757                         dst_port = portid;
 758
 759                 /* 02:00:00:00:00:xx */
 760                 d_addr_bytes = &eth_hdr->dst_addr.addr_bytes[0];
 761                 *((uint64_t *)d_addr_bytes) =
 762                         0x000000000002 + ((uint64_t)dst_port << 40);
 763
 764                 /* src addr */
 765                 rte_ether_addr_copy(&ports_eth_addr[dst_port],
 766                                 &eth_hdr->src_addr);
 767
 768                 send_single_packet(m, dst_port);
 769 #else
 770                 /* We don't currently handle IPv6 packets in LPM mode. */
 771                 rte_pktmbuf_free(m);
 772 #endif
 773         } else
 774                 rte_pktmbuf_free(m);
 775
 776 }
 777
 778 #define MINIMUM_SLEEP_TIME         1
 779 #define SUSPEND_THRESHOLD          300
 780
 781 static inline uint32_t
 782 power_idle_heuristic(uint32_t zero_rx_packet_count)
 783 {
 784         /* If zero count is less than 100,  sleep 1us */
 785         if (zero_rx_packet_count < SUSPEND_THRESHOLD)
 786                 return MINIMUM_SLEEP_TIME;
 787         /* If zero count is less than 1000, sleep 100 us which is the
 788                 minimum latency switching from C3/C6 to C0
 789         */
 790         else
 791                 return SUSPEND_THRESHOLD;
 792 }
 793
 794 static inline enum freq_scale_hint_t
 795 power_freq_scaleup_heuristic(unsigned lcore_id,
 796                              uint16_t port_id,
 797                              uint16_t queue_id)
 798 {
 799         uint32_t rxq_count = rte_eth_rx_queue_count(port_id, queue_id);
 800 /**
 801  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 802  * per iteration
 803  */
 804 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 805 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 806 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 807 #define FREQ_UP_TREND1_ACC   1
 808 #define FREQ_UP_TREND2_ACC   100
 809 #define FREQ_UP_THRESHOLD    10000
 810
 811         if (likely(rxq_count > FREQ_GEAR3_RX_PACKET_THRESHOLD)) {
 812                 stats[lcore_id].trend = 0;
 813                 return FREQ_HIGHEST;
 814         } else if (likely(rxq_count > FREQ_GEAR2_RX_PACKET_THRESHOLD))
 815                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 816         else if (likely(rxq_count > FREQ_GEAR1_RX_PACKET_THRESHOLD))
 817                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 818
 819         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 820                 stats[lcore_id].trend = 0;
 821                 return FREQ_HIGHER;
 822         }
 823
 824         return FREQ_CURRENT;
 825 }
 826
 827 /**
 828  * force polling thread sleep until one-shot rx interrupt triggers
 829  * @param port_id
 830  *  Port id.
 831  * @param queue_id
 832  *  Rx queue id.
 833  * @return
 834  *  0 on success
 835  */
 836 static int
 837 sleep_until_rx_interrupt(int num, int lcore)
 838 {
 839         /*
 840          * we want to track when we are woken up by traffic so that we can go
 841          * back to sleep again without log spamming. Avoid cache line sharing
 842          * to prevent threads stepping on each others' toes.
 843          */
 844         static struct {
 845                 bool wakeup;
 846         } __rte_cache_aligned status[RTE_MAX_LCORE];
 847         struct rte_epoll_event event[num];
 848         int n, i;
 849         uint16_t port_id;
 850         uint8_t queue_id;
 851         void *data;
 852
 853         if (status[lcore].wakeup) {
 854                 RTE_LOG(INFO, L3FWD_POWER,
 855                                 "lcore %u sleeps until interrupt triggers\n",
 856                                 rte_lcore_id());
 857         }
 858
 859         n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, 10);
 860         for (i = 0; i < n; i++) {
 861                 data = event[i].epdata.data;
 862                 port_id = ((uintptr_t)data) >> CHAR_BIT;
 863                 queue_id = ((uintptr_t)data) &
 864                         RTE_LEN2MASK(CHAR_BIT, uint8_t);
 865                 RTE_LOG(INFO, L3FWD_POWER,
 866                         "lcore %u is waked up from rx interrupt on"
 867                         " port %d queue %d\n",
 868                         rte_lcore_id(), port_id, queue_id);
 869         }
 870         status[lcore].wakeup = n != 0;
 871
 872         return 0;
 873 }
 874
 875 static void turn_on_off_intr(struct lcore_conf *qconf, bool on)
 876 {
 877         int i;
 878         struct lcore_rx_queue *rx_queue;
 879         uint8_t queue_id;
 880         uint16_t port_id;
 881
 882         for (i = 0; i < qconf->n_rx_queue; ++i) {
 883                 rx_queue = &(qconf->rx_queue_list[i]);
 884                 port_id = rx_queue->port_id;
 885                 queue_id = rx_queue->queue_id;
 886
 887                 rte_spinlock_lock(&(locks[port_id]));
 888                 if (on)
 889                         rte_eth_dev_rx_intr_enable(port_id, queue_id);
 890                 else
 891                         rte_eth_dev_rx_intr_disable(port_id, queue_id);
 892                 rte_spinlock_unlock(&(locks[port_id]));
 893         }
 894 }
 895
 896 static int event_register(struct lcore_conf *qconf)
 897 {
 898         struct lcore_rx_queue *rx_queue;
 899         uint8_t queueid;
 900         uint16_t portid;
 901         uint32_t data;
 902         int ret;
 903         int i;
 904
 905         for (i = 0; i < qconf->n_rx_queue; ++i) {
 906                 rx_queue = &(qconf->rx_queue_list[i]);
 907                 portid = rx_queue->port_id;
 908                 queueid = rx_queue->queue_id;
 909                 data = portid << CHAR_BIT | queueid;
 910
 911                 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
 912                                                 RTE_EPOLL_PER_THREAD,
 913                                                 RTE_INTR_EVENT_ADD,
 914                                                 (void *)((uintptr_t)data));
 915                 if (ret)
 916                         return ret;
 917         }
 918
 919         return 0;
 920 }
 921
 922 /* Main processing loop. 8< */
 923 static int main_intr_loop(__rte_unused void *dummy)
 924 {
 925         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 926         unsigned int lcore_id;
 927         uint64_t prev_tsc, diff_tsc, cur_tsc;
 928         int i, j, nb_rx;
 929         uint8_t queueid;
 930         uint16_t portid;
 931         struct lcore_conf *qconf;
 932         struct lcore_rx_queue *rx_queue;
 933         uint32_t lcore_rx_idle_count = 0;
 934         uint32_t lcore_idle_hint = 0;
 935         int intr_en = 0;
 936
 937         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
 938                                    US_PER_S * BURST_TX_DRAIN_US;
 939
 940         prev_tsc = 0;
 941
 942         lcore_id = rte_lcore_id();
 943         qconf = &lcore_conf[lcore_id];
 944
 945         if (qconf->n_rx_queue == 0) {
 946                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
 947                                 lcore_id);
 948                 return 0;
 949         }
 950
 951         RTE_LOG(INFO, L3FWD_POWER, "entering main interrupt loop on lcore %u\n",
 952                         lcore_id);
 953
 954         for (i = 0; i < qconf->n_rx_queue; i++) {
 955                 portid = qconf->rx_queue_list[i].port_id;
 956                 queueid = qconf->rx_queue_list[i].queue_id;
 957                 RTE_LOG(INFO, L3FWD_POWER,
 958                                 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
 959                                 lcore_id, portid, queueid);
 960         }
 961
 962         /* add into event wait list */
 963         if (event_register(qconf) == 0)
 964                 intr_en = 1;
 965         else
 966                 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
 967
 968         while (!is_done()) {
 969                 stats[lcore_id].nb_iteration_looped++;
 970
 971                 cur_tsc = rte_rdtsc();
 972
 973                 /*
 974                  * TX burst queue drain
 975                  */
 976                 diff_tsc = cur_tsc - prev_tsc;
 977                 if (unlikely(diff_tsc > drain_tsc)) {
 978                         for (i = 0; i < qconf->n_tx_port; ++i) {
 979                                 portid = qconf->tx_port_id[i];
 980                                 rte_eth_tx_buffer_flush(portid,
 981                                                 qconf->tx_queue_id[portid],
 982                                                 qconf->tx_buffer[portid]);
 983                         }
 984                         prev_tsc = cur_tsc;
 985                 }
 986
 987 start_rx:
 988                 /*
 989                  * Read packet from RX queues
 990                  */
 991                 lcore_rx_idle_count = 0;
 992                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 993                         rx_queue = &(qconf->rx_queue_list[i]);
 994                         rx_queue->idle_hint = 0;
 995                         portid = rx_queue->port_id;
 996                         queueid = rx_queue->queue_id;
 997
 998                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 999                                         MAX_PKT_BURST);
1000
1001                         stats[lcore_id].nb_rx_processed += nb_rx;
1002                         if (unlikely(nb_rx == 0)) {
1003                                 /**
1004                                  * no packet received from rx queue, try to
1005                                  * sleep for a while forcing CPU enter deeper
1006                                  * C states.
1007                                  */
1008                                 rx_queue->zero_rx_packet_count++;
1009
1010                                 if (rx_queue->zero_rx_packet_count <=
1011                                                 MIN_ZERO_POLL_COUNT)
1012                                         continue;
1013
1014                                 rx_queue->idle_hint = power_idle_heuristic(
1015                                                 rx_queue->zero_rx_packet_count);
1016                                 lcore_rx_idle_count++;
1017                         } else {
1018                                 rx_queue->zero_rx_packet_count = 0;
1019                         }
1020
1021                         /* Prefetch first packets */
1022                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1023                                 rte_prefetch0(rte_pktmbuf_mtod(
1024                                                 pkts_burst[j], void *));
1025                         }
1026
1027                         /* Prefetch and forward already prefetched packets */
1028                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1029                                 rte_prefetch0(rte_pktmbuf_mtod(
1030                                                 pkts_burst[j + PREFETCH_OFFSET],
1031                                                 void *));
1032                                 l3fwd_simple_forward(
1033                                                 pkts_burst[j], portid, qconf);
1034                         }
1035
1036                         /* Forward remaining prefetched packets */
1037                         for (; j < nb_rx; j++) {
1038                                 l3fwd_simple_forward(
1039                                                 pkts_burst[j], portid, qconf);
1040                         }
1041                 }
1042
1043                 if (unlikely(lcore_rx_idle_count == qconf->n_rx_queue)) {
1044                         /**
1045                          * All Rx queues empty in recent consecutive polls,
1046                          * sleep in a conservative manner, meaning sleep as
1047                          * less as possible.
1048                          */
1049                         for (i = 1,
1050                             lcore_idle_hint = qconf->rx_queue_list[0].idle_hint;
1051                                         i < qconf->n_rx_queue; ++i) {
1052                                 rx_queue = &(qconf->rx_queue_list[i]);
1053                                 if (rx_queue->idle_hint < lcore_idle_hint)
1054                                         lcore_idle_hint = rx_queue->idle_hint;
1055                         }
1056
1057                         if (lcore_idle_hint < SUSPEND_THRESHOLD)
1058                                 /**
1059                                  * execute "pause" instruction to avoid context
1060                                  * switch which generally take hundred of
1061                                  * microseconds for short sleep.
1062                                  */
1063                                 rte_delay_us(lcore_idle_hint);
1064                         else {
1065                                 /* suspend until rx interrupt triggers */
1066                                 if (intr_en) {
1067                                         turn_on_off_intr(qconf, 1);
1068                                         sleep_until_rx_interrupt(
1069                                                         qconf->n_rx_queue,
1070                                                         lcore_id);
1071                                         turn_on_off_intr(qconf, 0);
1072                                         /**
1073                                          * start receiving packets immediately
1074                                          */
1075                                         if (likely(!is_done()))
1076                                                 goto start_rx;
1077                                 }
1078                         }
1079                         stats[lcore_id].sleep_time += lcore_idle_hint;
1080                 }
1081         }
1082
1083         return 0;
1084 }
1085 /* >8 End of main processing loop. */
1086
1087 /* main processing loop */
1088 static int
1089 main_telemetry_loop(__rte_unused void *dummy)
1090 {
1091         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1092         unsigned int lcore_id;
1093         uint64_t prev_tsc, diff_tsc, cur_tsc, prev_tel_tsc;
1094         int i, j, nb_rx;
1095         uint8_t queueid;
1096         uint16_t portid;
1097         struct lcore_conf *qconf;
1098         struct lcore_rx_queue *rx_queue;
1099         uint64_t ep_nep[2] = {0}, fp_nfp[2] = {0};
1100         uint64_t poll_count;
1101         enum busy_rate br;
1102
1103         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
1104                                         US_PER_S * BURST_TX_DRAIN_US;
1105
1106         poll_count = 0;
1107         prev_tsc = 0;
1108         prev_tel_tsc = 0;
1109
1110         lcore_id = rte_lcore_id();
1111         qconf = &lcore_conf[lcore_id];
1112
1113         if (qconf->n_rx_queue == 0) {
1114                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
1115                         lcore_id);
1116                 return 0;
1117         }
1118
1119         RTE_LOG(INFO, L3FWD_POWER, "entering main telemetry loop on lcore %u\n",
1120                 lcore_id);
1121
1122         for (i = 0; i < qconf->n_rx_queue; i++) {
1123                 portid = qconf->rx_queue_list[i].port_id;
1124                 queueid = qconf->rx_queue_list[i].queue_id;
1125                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1126                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1127         }
1128
1129         while (!is_done()) {
1130
1131                 cur_tsc = rte_rdtsc();
1132                 /*
1133                  * TX burst queue drain
1134                  */
1135                 diff_tsc = cur_tsc - prev_tsc;
1136                 if (unlikely(diff_tsc > drain_tsc)) {
1137                         for (i = 0; i < qconf->n_tx_port; ++i) {
1138                                 portid = qconf->tx_port_id[i];
1139                                 rte_eth_tx_buffer_flush(portid,
1140                                                 qconf->tx_queue_id[portid],
1141                                                 qconf->tx_buffer[portid]);
1142                         }
1143                         prev_tsc = cur_tsc;
1144                 }
1145
1146                 /*
1147                  * Read packet from RX queues
1148                  */
1149                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1150                         rx_queue = &(qconf->rx_queue_list[i]);
1151                         portid = rx_queue->port_id;
1152                         queueid = rx_queue->queue_id;
1153
1154                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1155                                                                 MAX_PKT_BURST);
1156                         ep_nep[nb_rx == 0]++;
1157                         fp_nfp[nb_rx == MAX_PKT_BURST]++;
1158                         poll_count++;
1159                         RTE_LOG(INFO, L3FWD_POWER,
1160                                 "lcore %u has woken up on port %d queue %d\n",
1161                                 rte_lcore_id(), portid, queueid);
1162                         if (unlikely(nb_rx == 0))
1163                                 continue;
1164
1165                         /* Prefetch first packets */
1166                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1167                                 rte_prefetch0(rte_pktmbuf_mtod(
1168                                                 pkts_burst[j], void *));
1169                         }
1170
1171                         /* Prefetch and forward already prefetched packets */
1172                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1173                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1174                                                 j + PREFETCH_OFFSET], void *));
1175                                 l3fwd_simple_forward(pkts_burst[j], portid,
1176                                                                 qconf);
1177                         }
1178
1179                         /* Forward remaining prefetched packets */
1180                         for (; j < nb_rx; j++) {
1181                                 l3fwd_simple_forward(pkts_burst[j], portid,
1182                                                                 qconf);
1183                         }
1184                 }
1185                 if (unlikely(poll_count >= DEFAULT_COUNT)) {
1186                         diff_tsc = cur_tsc - prev_tel_tsc;
1187                         if (diff_tsc >= MAX_CYCLES) {
1188                                 br = FULL;
1189                         } else if (diff_tsc > MIN_CYCLES &&
1190                                         diff_tsc < MAX_CYCLES) {
1191                                 br = (diff_tsc * 100) / MAX_CYCLES;
1192                         } else {
1193                                 br = ZERO;
1194                         }
1195                         poll_count = 0;
1196                         prev_tel_tsc = cur_tsc;
1197                         /* update stats for telemetry */
1198                         rte_spinlock_lock(&stats[lcore_id].telemetry_lock);
1199                         stats[lcore_id].ep_nep[0] = ep_nep[0];
1200                         stats[lcore_id].ep_nep[1] = ep_nep[1];
1201                         stats[lcore_id].fp_nfp[0] = fp_nfp[0];
1202                         stats[lcore_id].fp_nfp[1] = fp_nfp[1];
1203                         stats[lcore_id].br = br;
1204                         rte_spinlock_unlock(&stats[lcore_id].telemetry_lock);
1205                 }
1206         }
1207
1208         return 0;
1209 }
1210 /* main processing loop */
1211 static int
1212 main_empty_poll_loop(__rte_unused void *dummy)
1213 {
1214         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1215         unsigned int lcore_id;
1216         uint64_t prev_tsc, diff_tsc, cur_tsc;
1217         int i, j, nb_rx;
1218         uint8_t queueid;
1219         uint16_t portid;
1220         struct lcore_conf *qconf;
1221         struct lcore_rx_queue *rx_queue;
1222
1223         const uint64_t drain_tsc =
1224                 (rte_get_tsc_hz() + US_PER_S - 1) /
1225                 US_PER_S * BURST_TX_DRAIN_US;
1226
1227         prev_tsc = 0;
1228
1229         lcore_id = rte_lcore_id();
1230         qconf = &lcore_conf[lcore_id];
1231
1232         if (qconf->n_rx_queue == 0) {
1233                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n",
1234                         lcore_id);
1235                 return 0;
1236         }
1237
1238         for (i = 0; i < qconf->n_rx_queue; i++) {
1239                 portid = qconf->rx_queue_list[i].port_id;
1240                 queueid = qconf->rx_queue_list[i].queue_id;
1241                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1242                                 "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1243         }
1244
1245         while (!is_done()) {
1246                 stats[lcore_id].nb_iteration_looped++;
1247
1248                 cur_tsc = rte_rdtsc();
1249                 /*
1250                  * TX burst queue drain
1251                  */
1252                 diff_tsc = cur_tsc - prev_tsc;
1253                 if (unlikely(diff_tsc > drain_tsc)) {
1254                         for (i = 0; i < qconf->n_tx_port; ++i) {
1255                                 portid = qconf->tx_port_id[i];
1256                                 rte_eth_tx_buffer_flush(portid,
1257                                                 qconf->tx_queue_id[portid],
1258                                                 qconf->tx_buffer[portid]);
1259                         }
1260                         prev_tsc = cur_tsc;
1261                 }
1262
1263                 /*
1264                  * Read packet from RX queues
1265                  */
1266                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1267                         rx_queue = &(qconf->rx_queue_list[i]);
1268                         rx_queue->idle_hint = 0;
1269                         portid = rx_queue->port_id;
1270                         queueid = rx_queue->queue_id;
1271
1272                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1273                                         MAX_PKT_BURST);
1274
1275                         stats[lcore_id].nb_rx_processed += nb_rx;
1276
1277                         if (nb_rx == 0) {
1278
1279                                 rte_power_empty_poll_stat_update(lcore_id);
1280
1281                                 continue;
1282                         } else {
1283                                 rte_power_poll_stat_update(lcore_id, nb_rx);
1284                         }
1285
1286
1287                         /* Prefetch first packets */
1288                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1289                                 rte_prefetch0(rte_pktmbuf_mtod(
1290                                                         pkts_burst[j], void *));
1291                         }
1292
1293                         /* Prefetch and forward already prefetched packets */
1294                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1295                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1296                                                         j + PREFETCH_OFFSET],
1297                                                         void *));
1298                                 l3fwd_simple_forward(pkts_burst[j], portid,
1299                                                 qconf);
1300                         }
1301
1302                         /* Forward remaining prefetched packets */
1303                         for (; j < nb_rx; j++) {
1304                                 l3fwd_simple_forward(pkts_burst[j], portid,
1305                                                 qconf);
1306                         }
1307
1308                 }
1309
1310         }
1311
1312         return 0;
1313 }
1314 /* main processing loop */
1315 static int
1316 main_legacy_loop(__rte_unused void *dummy)
1317 {
1318         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1319         unsigned lcore_id;
1320         uint64_t prev_tsc, diff_tsc, cur_tsc, tim_res_tsc, hz;
1321         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
1322         int i, j, nb_rx;
1323         uint8_t queueid;
1324         uint16_t portid;
1325         struct lcore_conf *qconf;
1326         struct lcore_rx_queue *rx_queue;
1327         enum freq_scale_hint_t lcore_scaleup_hint;
1328         uint32_t lcore_rx_idle_count = 0;
1329         uint32_t lcore_idle_hint = 0;
1330         int intr_en = 0;
1331
1332         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
1333
1334         prev_tsc = 0;
1335         hz = rte_get_timer_hz();
1336         tim_res_tsc = hz/TIMER_NUMBER_PER_SECOND;
1337
1338         lcore_id = rte_lcore_id();
1339         qconf = &lcore_conf[lcore_id];
1340
1341         if (qconf->n_rx_queue == 0) {
1342                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
1343                 return 0;
1344         }
1345
1346         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
1347
1348         for (i = 0; i < qconf->n_rx_queue; i++) {
1349                 portid = qconf->rx_queue_list[i].port_id;
1350                 queueid = qconf->rx_queue_list[i].queue_id;
1351                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
1352                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
1353         }
1354
1355         /* add into event wait list */
1356         if (event_register(qconf) == 0)
1357                 intr_en = 1;
1358         else
1359                 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
1360
1361         while (!is_done()) {
1362                 stats[lcore_id].nb_iteration_looped++;
1363
1364                 cur_tsc = rte_rdtsc();
1365                 cur_tsc_power = cur_tsc;
1366
1367                 /*
1368                  * TX burst queue drain
1369                  */
1370                 diff_tsc = cur_tsc - prev_tsc;
1371                 if (unlikely(diff_tsc > drain_tsc)) {
1372                         for (i = 0; i < qconf->n_tx_port; ++i) {
1373                                 portid = qconf->tx_port_id[i];
1374                                 rte_eth_tx_buffer_flush(portid,
1375                                                 qconf->tx_queue_id[portid],
1376                                                 qconf->tx_buffer[portid]);
1377                         }
1378                         prev_tsc = cur_tsc;
1379                 }
1380
1381                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
1382                 if (diff_tsc_power > tim_res_tsc) {
1383                         rte_timer_manage();
1384                         prev_tsc_power = cur_tsc_power;
1385                 }
1386
1387 start_rx:
1388                 /*
1389                  * Read packet from RX queues
1390                  */
1391                 lcore_scaleup_hint = FREQ_CURRENT;
1392                 lcore_rx_idle_count = 0;
1393                 for (i = 0; i < qconf->n_rx_queue; ++i) {
1394                         rx_queue = &(qconf->rx_queue_list[i]);
1395                         rx_queue->idle_hint = 0;
1396                         portid = rx_queue->port_id;
1397                         queueid = rx_queue->queue_id;
1398
1399                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
1400                                                                 MAX_PKT_BURST);
1401
1402                         stats[lcore_id].nb_rx_processed += nb_rx;
1403                         if (unlikely(nb_rx == 0)) {
1404                                 /**
1405                                  * no packet received from rx queue, try to
1406                                  * sleep for a while forcing CPU enter deeper
1407                                  * C states.
1408                                  */
1409                                 rx_queue->zero_rx_packet_count++;
1410
1411                                 if (rx_queue->zero_rx_packet_count <=
1412                                                         MIN_ZERO_POLL_COUNT)
1413                                         continue;
1414
1415                                 rx_queue->idle_hint = power_idle_heuristic(\
1416                                         rx_queue->zero_rx_packet_count);
1417                                 lcore_rx_idle_count++;
1418                         } else {
1419                                 rx_queue->zero_rx_packet_count = 0;
1420
1421                                 /**
1422                                  * do not scale up frequency immediately as
1423                                  * user to kernel space communication is costly
1424                                  * which might impact packet I/O for received
1425                                  * packets.
1426                                  */
1427                                 rx_queue->freq_up_hint =
1428                                         power_freq_scaleup_heuristic(lcore_id,
1429                                                         portid, queueid);
1430                         }
1431
1432                         /* Prefetch first packets */
1433                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
1434                                 rte_prefetch0(rte_pktmbuf_mtod(
1435                                                 pkts_burst[j], void *));
1436                         }
1437
1438                         /* Prefetch and forward already prefetched packets */
1439                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
1440                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
1441                                                 j + PREFETCH_OFFSET], void *));
1442                                 l3fwd_simple_forward(pkts_burst[j], portid,
1443                                                                 qconf);
1444                         }
1445
1446                         /* Forward remaining prefetched packets */
1447                         for (; j < nb_rx; j++) {
1448                                 l3fwd_simple_forward(pkts_burst[j], portid,
1449                                                                 qconf);
1450                         }
1451                 }
1452
1453                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
1454                         for (i = 1, lcore_scaleup_hint =
1455                                 qconf->rx_queue_list[0].freq_up_hint;
1456                                         i < qconf->n_rx_queue; ++i) {
1457                                 rx_queue = &(qconf->rx_queue_list[i]);
1458                                 if (rx_queue->freq_up_hint >
1459                                                 lcore_scaleup_hint)
1460                                         lcore_scaleup_hint =
1461                                                 rx_queue->freq_up_hint;
1462                         }
1463
1464                         if (lcore_scaleup_hint == FREQ_HIGHEST) {
1465                                 if (rte_power_freq_max)
1466                                         rte_power_freq_max(lcore_id);
1467                         } else if (lcore_scaleup_hint == FREQ_HIGHER) {
1468                                 if (rte_power_freq_up)
1469                                         rte_power_freq_up(lcore_id);
1470                         }
1471                 } else {
1472                         /**
1473                          * All Rx queues empty in recent consecutive polls,
1474                          * sleep in a conservative manner, meaning sleep as
1475                          * less as possible.
1476                          */
1477                         for (i = 1, lcore_idle_hint =
1478                                 qconf->rx_queue_list[0].idle_hint;
1479                                         i < qconf->n_rx_queue; ++i) {
1480                                 rx_queue = &(qconf->rx_queue_list[i]);
1481                                 if (rx_queue->idle_hint < lcore_idle_hint)
1482                                         lcore_idle_hint = rx_queue->idle_hint;
1483                         }
1484
1485                         if (lcore_idle_hint < SUSPEND_THRESHOLD)
1486                                 /**
1487                                  * execute "pause" instruction to avoid context
1488                                  * switch which generally take hundred of
1489                                  * microseconds for short sleep.
1490                                  */
1491                                 rte_delay_us(lcore_idle_hint);
1492                         else {
1493                                 /* suspend until rx interrupt triggers */
1494                                 if (intr_en) {
1495                                         turn_on_off_intr(qconf, 1);
1496                                         sleep_until_rx_interrupt(
1497                                                         qconf->n_rx_queue,
1498                                                         lcore_id);
1499                                         turn_on_off_intr(qconf, 0);
1500                                         /**
1501                                          * start receiving packets immediately
1502                                          */
1503                                         if (likely(!is_done()))
1504                                                 goto start_rx;
1505                                 }
1506                         }
1507                         stats[lcore_id].sleep_time += lcore_idle_hint;
1508                 }
1509         }
1510
1511         return 0;
1512 }
1513
1514 static int
1515 check_lcore_params(void)
1516 {
1517         uint8_t queue, lcore;
1518         uint16_t i;
1519         int socketid;
1520
1521         for (i = 0; i < nb_lcore_params; ++i) {
1522                 queue = lcore_params[i].queue_id;
1523                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1524                         printf("invalid queue number: %hhu\n", queue);
1525                         return -1;
1526                 }
1527                 lcore = lcore_params[i].lcore_id;
1528                 if (!rte_lcore_is_enabled(lcore)) {
1529                         printf("error: lcore %hhu is not enabled in lcore "
1530                                                         "mask\n", lcore);
1531                         return -1;
1532                 }
1533                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1534                                                         (numa_on == 0)) {
1535                         printf("warning: lcore %hhu is on socket %d with numa "
1536                                                 "off\n", lcore, socketid);
1537                 }
1538                 if (app_mode == APP_MODE_TELEMETRY && lcore == rte_lcore_id()) {
1539                         printf("cannot enable main core %d in config for telemetry mode\n",
1540                                 rte_lcore_id());
1541                         return -1;
1542                 }
1543         }
1544         return 0;
1545 }
1546
1547 static int
1548 check_port_config(void)
1549 {
1550         unsigned portid;
1551         uint16_t i;
1552
1553         for (i = 0; i < nb_lcore_params; ++i) {
1554                 portid = lcore_params[i].port_id;
1555                 if ((enabled_port_mask & (1 << portid)) == 0) {
1556                         printf("port %u is not enabled in port mask\n",
1557                                                                 portid);
1558                         return -1;
1559                 }
1560                 if (!rte_eth_dev_is_valid_port(portid)) {
1561                         printf("port %u is not present on the board\n",
1562                                                                 portid);
1563                         return -1;
1564                 }
1565         }
1566         return 0;
1567 }
1568
1569 static uint8_t
1570 get_port_n_rx_queues(const uint16_t port)
1571 {
1572         int queue = -1;
1573         uint16_t i;
1574
1575         for (i = 0; i < nb_lcore_params; ++i) {
1576                 if (lcore_params[i].port_id == port &&
1577                                 lcore_params[i].queue_id > queue)
1578                         queue = lcore_params[i].queue_id;
1579         }
1580         return (uint8_t)(++queue);
1581 }
1582
1583 static int
1584 init_lcore_rx_queues(void)
1585 {
1586         uint16_t i, nb_rx_queue;
1587         uint8_t lcore;
1588
1589         for (i = 0; i < nb_lcore_params; ++i) {
1590                 lcore = lcore_params[i].lcore_id;
1591                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1592                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1593                         printf("error: too many queues (%u) for lcore: %u\n",
1594                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1595                         return -1;
1596                 } else {
1597                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1598                                 lcore_params[i].port_id;
1599                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1600                                 lcore_params[i].queue_id;
1601                         lcore_conf[lcore].n_rx_queue++;
1602                 }
1603         }
1604         return 0;
1605 }
1606
1607 /* display usage */
1608 static void
1609 print_usage(const char *prgname)
1610 {
1611         printf ("%s [EAL options] -- -p PORTMASK -P"
1612                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1613                 "  [--high-perf-cores CORELIST"
1614                 "  [--perf-config (port,queue,hi_perf,lcore_index)[,(port,queue,hi_perf,lcore_index]]"
1615                 "  [--max-pkt-len PKTLEN]\n"
1616                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1617                 "  -P: enable promiscuous mode\n"
1618                 "  --config (port,queue,lcore): rx queues configuration\n"
1619                 "  --high-perf-cores CORELIST: list of high performance cores\n"
1620                 "  --perf-config: similar as config, cores specified as indices"
1621                 " for bins containing high or regular performance cores\n"
1622                 "  --no-numa: optional, disable numa awareness\n"
1623                 "  --max-pkt-len PKTLEN: maximum packet length in decimal (64-9600)\n"
1624                 "  --parse-ptype: parse packet type by software\n"
1625                 "  --legacy: use legacy interrupt-based scaling\n"
1626                 "  --empty-poll: enable empty poll detection"
1627                 " follow (training_flag, high_threshold, med_threshold)\n"
1628                 " --telemetry: enable telemetry mode, to update"
1629                 " empty polls, full polls, and core busyness to telemetry\n"
1630                 " --interrupt-only: enable interrupt-only mode\n"
1631                 " --pmd-mgmt MODE: enable PMD power management mode. "
1632                 "Currently supported modes: baseline, monitor, pause, scale\n",
1633                 prgname);
1634 }
1635
1636 static int parse_max_pkt_len(const char *pktlen)
1637 {
1638         char *end = NULL;
1639         unsigned long len;
1640
1641         /* parse decimal string */
1642         len = strtoul(pktlen, &end, 10);
1643         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1644                 return -1;
1645
1646         if (len == 0)
1647                 return -1;
1648
1649         return len;
1650 }
1651
1652 static int
1653 parse_portmask(const char *portmask)
1654 {
1655         char *end = NULL;
1656         unsigned long pm;
1657
1658         /* parse hexadecimal string */
1659         pm = strtoul(portmask, &end, 16);
1660         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1661                 return 0;
1662
1663         return pm;
1664 }
1665
1666 static int
1667 parse_config(const char *q_arg)
1668 {
1669         char s[256];
1670         const char *p, *p0 = q_arg;
1671         char *end;
1672         enum fieldnames {
1673                 FLD_PORT = 0,
1674                 FLD_QUEUE,
1675                 FLD_LCORE,
1676                 _NUM_FLD
1677         };
1678         unsigned long int_fld[_NUM_FLD];
1679         char *str_fld[_NUM_FLD];
1680         int i;
1681         unsigned size;
1682
1683         nb_lcore_params = 0;
1684
1685         while ((p = strchr(p0,'(')) != NULL) {
1686                 ++p;
1687                 if((p0 = strchr(p,')')) == NULL)
1688                         return -1;
1689
1690                 size = p0 - p;
1691                 if(size >= sizeof(s))
1692                         return -1;
1693
1694                 snprintf(s, sizeof(s), "%.*s", size, p);
1695                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1696                                                                 _NUM_FLD)
1697                         return -1;
1698                 for (i = 0; i < _NUM_FLD; i++){
1699                         errno = 0;
1700                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1701                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1702                                                                         255)
1703                                 return -1;
1704                 }
1705                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1706                         printf("exceeded max number of lcore params: %hu\n",
1707                                 nb_lcore_params);
1708                         return -1;
1709                 }
1710                 lcore_params_array[nb_lcore_params].port_id =
1711                                 (uint8_t)int_fld[FLD_PORT];
1712                 lcore_params_array[nb_lcore_params].queue_id =
1713                                 (uint8_t)int_fld[FLD_QUEUE];
1714                 lcore_params_array[nb_lcore_params].lcore_id =
1715                                 (uint8_t)int_fld[FLD_LCORE];
1716                 ++nb_lcore_params;
1717         }
1718         lcore_params = lcore_params_array;
1719
1720         return 0;
1721 }
1722
1723 static int
1724 parse_pmd_mgmt_config(const char *name)
1725 {
1726 #define PMD_MGMT_MONITOR "monitor"
1727 #define PMD_MGMT_PAUSE   "pause"
1728 #define PMD_MGMT_SCALE   "scale"
1729 #define PMD_MGMT_BASELINE  "baseline"
1730
1731         if (strncmp(PMD_MGMT_MONITOR, name, sizeof(PMD_MGMT_MONITOR)) == 0) {
1732                 pmgmt_type = RTE_POWER_MGMT_TYPE_MONITOR;
1733                 return 0;
1734         }
1735
1736         if (strncmp(PMD_MGMT_PAUSE, name, sizeof(PMD_MGMT_PAUSE)) == 0) {
1737                 pmgmt_type = RTE_POWER_MGMT_TYPE_PAUSE;
1738                 return 0;
1739         }
1740
1741         if (strncmp(PMD_MGMT_SCALE, name, sizeof(PMD_MGMT_SCALE)) == 0) {
1742                 pmgmt_type = RTE_POWER_MGMT_TYPE_SCALE;
1743                 return 0;
1744         }
1745         if (strncmp(PMD_MGMT_BASELINE, name, sizeof(PMD_MGMT_BASELINE)) == 0) {
1746                 baseline_enabled = true;
1747                 return 0;
1748         }
1749         /* unknown PMD power management mode */
1750         return -1;
1751 }
1752
1753 static int
1754 parse_ep_config(const char *q_arg)
1755 {
1756         char s[256];
1757         const char *p = q_arg;
1758         char *end;
1759         int  num_arg;
1760
1761         char *str_fld[3];
1762
1763         int training_flag;
1764         int med_edpi;
1765         int hgh_edpi;
1766
1767         ep_med_edpi = EMPTY_POLL_MED_THRESHOLD;
1768         ep_hgh_edpi = EMPTY_POLL_HGH_THRESHOLD;
1769
1770         strlcpy(s, p, sizeof(s));
1771
1772         num_arg = rte_strsplit(s, sizeof(s), str_fld, 3, ',');
1773
1774         empty_poll_train = false;
1775
1776         if (num_arg == 0)
1777                 return 0;
1778
1779         if (num_arg == 3) {
1780
1781                 training_flag = strtoul(str_fld[0], &end, 0);
1782                 med_edpi = strtoul(str_fld[1], &end, 0);
1783                 hgh_edpi = strtoul(str_fld[2], &end, 0);
1784
1785                 if (training_flag == 1)
1786                         empty_poll_train = true;
1787
1788                 if (med_edpi > 0)
1789                         ep_med_edpi = med_edpi;
1790
1791                 if (hgh_edpi > 0)
1792                         ep_hgh_edpi = hgh_edpi;
1793
1794         } else {
1795
1796                 return -1;
1797         }
1798
1799         return 0;
1800
1801 }
1802 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1803 #define CMD_LINE_OPT_LEGACY "legacy"
1804 #define CMD_LINE_OPT_EMPTY_POLL "empty-poll"
1805 #define CMD_LINE_OPT_INTERRUPT_ONLY "interrupt-only"
1806 #define CMD_LINE_OPT_TELEMETRY "telemetry"
1807 #define CMD_LINE_OPT_PMD_MGMT "pmd-mgmt"
1808 #define CMD_LINE_OPT_MAX_PKT_LEN "max-pkt-len"
1809
1810 /* Parse the argument given in the command line of the application */
1811 static int
1812 parse_args(int argc, char **argv)
1813 {
1814         int opt, ret;
1815         char **argvopt;
1816         int option_index;
1817         uint32_t limit;
1818         char *prgname = argv[0];
1819         static struct option lgopts[] = {
1820                 {"config", 1, 0, 0},
1821                 {"perf-config", 1, 0, 0},
1822                 {"high-perf-cores", 1, 0, 0},
1823                 {"no-numa", 0, 0, 0},
1824                 {CMD_LINE_OPT_MAX_PKT_LEN, 1, 0, 0},
1825                 {CMD_LINE_OPT_EMPTY_POLL, 1, 0, 0},
1826                 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1827                 {CMD_LINE_OPT_LEGACY, 0, 0, 0},
1828                 {CMD_LINE_OPT_TELEMETRY, 0, 0, 0},
1829                 {CMD_LINE_OPT_INTERRUPT_ONLY, 0, 0, 0},
1830                 {CMD_LINE_OPT_PMD_MGMT, 1, 0, 0},
1831                 {NULL, 0, 0, 0}
1832         };
1833
1834         argvopt = argv;
1835
1836         while ((opt = getopt_long(argc, argvopt, "p:l:m:h:P",
1837                                 lgopts, &option_index)) != EOF) {
1838
1839                 switch (opt) {
1840                 /* portmask */
1841                 case 'p':
1842                         enabled_port_mask = parse_portmask(optarg);
1843                         if (enabled_port_mask == 0) {
1844                                 printf("invalid portmask\n");
1845                                 print_usage(prgname);
1846                                 return -1;
1847                         }
1848                         break;
1849                 case 'P':
1850                         printf("Promiscuous mode selected\n");
1851                         promiscuous_on = 1;
1852                         break;
1853                 case 'l':
1854                         limit = parse_max_pkt_len(optarg);
1855                         freq_tlb[LOW] = limit;
1856                         break;
1857                 case 'm':
1858                         limit = parse_max_pkt_len(optarg);
1859                         freq_tlb[MED] = limit;
1860                         break;
1861                 case 'h':
1862                         limit = parse_max_pkt_len(optarg);
1863                         freq_tlb[HGH] = limit;
1864                         break;
1865                 /* long options */
1866                 case 0:
1867                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1868                                 ret = parse_config(optarg);
1869                                 if (ret) {
1870                                         printf("invalid config\n");
1871                                         print_usage(prgname);
1872                                         return -1;
1873                                 }
1874                         }
1875
1876                         if (!strncmp(lgopts[option_index].name,
1877                                         "perf-config", 11)) {
1878                                 ret = parse_perf_config(optarg);
1879                                 if (ret) {
1880                                         printf("invalid perf-config\n");
1881                                         print_usage(prgname);
1882                                         return -1;
1883                                 }
1884                         }
1885
1886                         if (!strncmp(lgopts[option_index].name,
1887                                         "high-perf-cores", 15)) {
1888                                 ret = parse_perf_core_list(optarg);
1889                                 if (ret) {
1890                                         printf("invalid high-perf-cores\n");
1891                                         print_usage(prgname);
1892                                         return -1;
1893                                 }
1894                         }
1895
1896                         if (!strncmp(lgopts[option_index].name,
1897                                                 "no-numa", 7)) {
1898                                 printf("numa is disabled \n");
1899                                 numa_on = 0;
1900                         }
1901
1902                         if (!strncmp(lgopts[option_index].name,
1903                                         CMD_LINE_OPT_LEGACY,
1904                                         sizeof(CMD_LINE_OPT_LEGACY))) {
1905                                 if (app_mode != APP_MODE_DEFAULT) {
1906                                         printf(" legacy mode is mutually exclusive with other modes\n");
1907                                         return -1;
1908                                 }
1909                                 app_mode = APP_MODE_LEGACY;
1910                                 printf("legacy mode is enabled\n");
1911                         }
1912
1913                         if (!strncmp(lgopts[option_index].name,
1914                                         CMD_LINE_OPT_EMPTY_POLL, 10)) {
1915                                 if (app_mode != APP_MODE_DEFAULT) {
1916                                         printf(" empty-poll mode is mutually exclusive with other modes\n");
1917                                         return -1;
1918                                 }
1919                                 app_mode = APP_MODE_EMPTY_POLL;
1920                                 ret = parse_ep_config(optarg);
1921
1922                                 if (ret) {
1923                                         printf("invalid empty poll config\n");
1924                                         print_usage(prgname);
1925                                         return -1;
1926                                 }
1927                                 printf("empty-poll is enabled\n");
1928                         }
1929
1930                         if (!strncmp(lgopts[option_index].name,
1931                                         CMD_LINE_OPT_TELEMETRY,
1932                                         sizeof(CMD_LINE_OPT_TELEMETRY))) {
1933                                 if (app_mode != APP_MODE_DEFAULT) {
1934                                         printf(" telemetry mode is mutually exclusive with other modes\n");
1935                                         return -1;
1936                                 }
1937                                 app_mode = APP_MODE_TELEMETRY;
1938                                 printf("telemetry mode is enabled\n");
1939                         }
1940
1941                         if (!strncmp(lgopts[option_index].name,
1942                                         CMD_LINE_OPT_PMD_MGMT,
1943                                         sizeof(CMD_LINE_OPT_PMD_MGMT))) {
1944                                 if (app_mode != APP_MODE_DEFAULT) {
1945                                         printf(" power mgmt mode is mutually exclusive with other modes\n");
1946                                         return -1;
1947                                 }
1948                                 if (parse_pmd_mgmt_config(optarg) < 0) {
1949                                         printf(" Invalid PMD power management mode: %s\n",
1950                                                         optarg);
1951                                         return -1;
1952                                 }
1953                                 app_mode = APP_MODE_PMD_MGMT;
1954                                 printf("PMD power mgmt mode is enabled\n");
1955                         }
1956                         if (!strncmp(lgopts[option_index].name,
1957                                         CMD_LINE_OPT_INTERRUPT_ONLY,
1958                                         sizeof(CMD_LINE_OPT_INTERRUPT_ONLY))) {
1959                                 if (app_mode != APP_MODE_DEFAULT) {
1960                                         printf(" interrupt-only mode is mutually exclusive with other modes\n");
1961                                         return -1;
1962                                 }
1963                                 app_mode = APP_MODE_INTERRUPT;
1964                                 printf("interrupt-only mode is enabled\n");
1965                         }
1966
1967                         if (!strncmp(lgopts[option_index].name,
1968                                         CMD_LINE_OPT_MAX_PKT_LEN,
1969                                         sizeof(CMD_LINE_OPT_MAX_PKT_LEN))) {
1970                                 printf("Custom frame size is configured\n");
1971                                 max_pkt_len = parse_max_pkt_len(optarg);
1972                         }
1973
1974                         if (!strncmp(lgopts[option_index].name,
1975                                      CMD_LINE_OPT_PARSE_PTYPE,
1976                                      sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1977                                 printf("soft parse-ptype is enabled\n");
1978                                 parse_ptype = 1;
1979                         }
1980
1981                         break;
1982
1983                 default:
1984                         print_usage(prgname);
1985                         return -1;
1986                 }
1987         }
1988
1989         if (optind >= 0)
1990                 argv[optind-1] = prgname;
1991
1992         ret = optind-1;
1993         optind = 1; /* reset getopt lib */
1994         return ret;
1995 }
1996
1997 static void
1998 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1999 {
2000         char buf[RTE_ETHER_ADDR_FMT_SIZE];
2001         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
2002         printf("%s%s", name, buf);
2003 }
2004
2005 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2006 static void
2007 setup_hash(int socketid)
2008 {
2009         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
2010                 .name = NULL,
2011                 .entries = L3FWD_HASH_ENTRIES,
2012                 .key_len = sizeof(struct ipv4_5tuple),
2013                 .hash_func = DEFAULT_HASH_FUNC,
2014                 .hash_func_init_val = 0,
2015         };
2016
2017         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
2018                 .name = NULL,
2019                 .entries = L3FWD_HASH_ENTRIES,
2020                 .key_len = sizeof(struct ipv6_5tuple),
2021                 .hash_func = DEFAULT_HASH_FUNC,
2022                 .hash_func_init_val = 0,
2023         };
2024
2025         unsigned i;
2026         int ret;
2027         char s[64];
2028
2029         /* create ipv4 hash */
2030         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
2031         ipv4_l3fwd_hash_params.name = s;
2032         ipv4_l3fwd_hash_params.socket_id = socketid;
2033         ipv4_l3fwd_lookup_struct[socketid] =
2034                 rte_hash_create(&ipv4_l3fwd_hash_params);
2035         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2036                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2037                                 "socket %d\n", socketid);
2038
2039         /* create ipv6 hash */
2040         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
2041         ipv6_l3fwd_hash_params.name = s;
2042         ipv6_l3fwd_hash_params.socket_id = socketid;
2043         ipv6_l3fwd_lookup_struct[socketid] =
2044                 rte_hash_create(&ipv6_l3fwd_hash_params);
2045         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
2046                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
2047                                 "socket %d\n", socketid);
2048
2049
2050         /* populate the ipv4 hash */
2051         for (i = 0; i < RTE_DIM(ipv4_l3fwd_route_array); i++) {
2052                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
2053                                 (void *) &ipv4_l3fwd_route_array[i].key);
2054                 if (ret < 0) {
2055                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
2056                                 "l3fwd hash on socket %d\n", i, socketid);
2057                 }
2058                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
2059                 printf("Hash: Adding key\n");
2060                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
2061         }
2062
2063         /* populate the ipv6 hash */
2064         for (i = 0; i < RTE_DIM(ipv6_l3fwd_route_array); i++) {
2065                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
2066                                 (void *) &ipv6_l3fwd_route_array[i].key);
2067                 if (ret < 0) {
2068                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
2069                                 "l3fwd hash on socket %d\n", i, socketid);
2070                 }
2071                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
2072                 printf("Hash: Adding key\n");
2073                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
2074         }
2075 }
2076 #endif
2077
2078 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2079 static void
2080 setup_lpm(int socketid)
2081 {
2082         unsigned i;
2083         int ret;
2084         char s[64];
2085
2086         /* create the LPM table */
2087         struct rte_lpm_config lpm_ipv4_config;
2088
2089         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
2090         lpm_ipv4_config.number_tbl8s = 256;
2091         lpm_ipv4_config.flags = 0;
2092
2093         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
2094         ipv4_l3fwd_lookup_struct[socketid] =
2095                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
2096         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
2097                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
2098                                 " on socket %d\n", socketid);
2099
2100         /* populate the LPM table */
2101         for (i = 0; i < RTE_DIM(ipv4_l3fwd_route_array); i++) {
2102                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
2103                         ipv4_l3fwd_route_array[i].ip,
2104                         ipv4_l3fwd_route_array[i].depth,
2105                         ipv4_l3fwd_route_array[i].if_out);
2106
2107                 if (ret < 0) {
2108                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
2109                                 "l3fwd LPM table on socket %d\n",
2110                                 i, socketid);
2111                 }
2112
2113                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
2114                         (unsigned)ipv4_l3fwd_route_array[i].ip,
2115                         ipv4_l3fwd_route_array[i].depth,
2116                         ipv4_l3fwd_route_array[i].if_out);
2117         }
2118 }
2119 #endif
2120
2121 static int
2122 init_mem(unsigned nb_mbuf)
2123 {
2124         struct lcore_conf *qconf;
2125         int socketid;
2126         unsigned lcore_id;
2127         char s[64];
2128
2129         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2130                 if (rte_lcore_is_enabled(lcore_id) == 0)
2131                         continue;
2132
2133                 if (numa_on)
2134                         socketid = rte_lcore_to_socket_id(lcore_id);
2135                 else
2136                         socketid = 0;
2137
2138                 if (socketid >= NB_SOCKETS) {
2139                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
2140                                         "out of range %d\n", socketid,
2141                                                 lcore_id, NB_SOCKETS);
2142                 }
2143                 if (pktmbuf_pool[socketid] == NULL) {
2144                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
2145                         pktmbuf_pool[socketid] =
2146                                 rte_pktmbuf_pool_create(s, nb_mbuf,
2147                                         MEMPOOL_CACHE_SIZE, 0,
2148                                         RTE_MBUF_DEFAULT_BUF_SIZE,
2149                                         socketid);
2150                         if (pktmbuf_pool[socketid] == NULL)
2151                                 rte_exit(EXIT_FAILURE,
2152                                         "Cannot init mbuf pool on socket %d\n",
2153                                                                 socketid);
2154                         else
2155                                 printf("Allocated mbuf pool on socket %d\n",
2156                                                                 socketid);
2157
2158 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2159                         setup_lpm(socketid);
2160 #else
2161                         setup_hash(socketid);
2162 #endif
2163                 }
2164                 qconf = &lcore_conf[lcore_id];
2165                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
2166 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2167                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
2168 #endif
2169         }
2170         return 0;
2171 }
2172
2173 /* Check the link status of all ports in up to 9s, and print them finally */
2174 static void
2175 check_all_ports_link_status(uint32_t port_mask)
2176 {
2177 #define CHECK_INTERVAL 100 /* 100ms */
2178 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
2179         uint8_t count, all_ports_up, print_flag = 0;
2180         uint16_t portid;
2181         struct rte_eth_link link;
2182         int ret;
2183         char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
2184
2185         printf("\nChecking link status");
2186         fflush(stdout);
2187         for (count = 0; count <= MAX_CHECK_TIME; count++) {
2188                 all_ports_up = 1;
2189                 RTE_ETH_FOREACH_DEV(portid) {
2190                         if ((port_mask & (1 << portid)) == 0)
2191                                 continue;
2192                         memset(&link, 0, sizeof(link));
2193                         ret = rte_eth_link_get_nowait(portid, &link);
2194                         if (ret < 0) {
2195                                 all_ports_up = 0;
2196                                 if (print_flag == 1)
2197                                         printf("Port %u link get failed: %s\n",
2198                                                 portid, rte_strerror(-ret));
2199                                 continue;
2200                         }
2201                         /* print link status if flag set */
2202                         if (print_flag == 1) {
2203                                 rte_eth_link_to_str(link_status_text,
2204                                         sizeof(link_status_text), &link);
2205                                 printf("Port %d %s\n", portid,
2206                                        link_status_text);
2207                                 continue;
2208                         }
2209                         /* clear all_ports_up flag if any link down */
2210                         if (link.link_status == RTE_ETH_LINK_DOWN) {
2211                                 all_ports_up = 0;
2212                                 break;
2213                         }
2214                 }
2215                 /* after finally printing all link status, get out */
2216                 if (print_flag == 1)
2217                         break;
2218
2219                 if (all_ports_up == 0) {
2220                         printf(".");
2221                         fflush(stdout);
2222                         rte_delay_ms(CHECK_INTERVAL);
2223                 }
2224
2225                 /* set the print_flag if all ports up or timeout */
2226                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2227                         print_flag = 1;
2228                         printf("done\n");
2229                 }
2230         }
2231 }
2232
2233 static int check_ptype(uint16_t portid)
2234 {
2235         int i, ret;
2236         int ptype_l3_ipv4 = 0;
2237 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2238         int ptype_l3_ipv6 = 0;
2239 #endif
2240         uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
2241
2242         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
2243         if (ret <= 0)
2244                 return 0;
2245
2246         uint32_t ptypes[ret];
2247
2248         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
2249         for (i = 0; i < ret; ++i) {
2250                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
2251                         ptype_l3_ipv4 = 1;
2252 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2253                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
2254                         ptype_l3_ipv6 = 1;
2255 #endif
2256         }
2257
2258         if (ptype_l3_ipv4 == 0)
2259                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
2260
2261 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
2262         if (ptype_l3_ipv6 == 0)
2263                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
2264 #endif
2265
2266 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
2267         if (ptype_l3_ipv4)
2268 #else /* APP_LOOKUP_EXACT_MATCH */
2269         if (ptype_l3_ipv4 && ptype_l3_ipv6)
2270 #endif
2271                 return 1;
2272
2273         return 0;
2274
2275 }
2276
2277 static int
2278 init_power_library(void)
2279 {
2280         enum power_management_env env;
2281         unsigned int lcore_id;
2282         int ret = 0;
2283
2284         RTE_LCORE_FOREACH(lcore_id) {
2285                 /* init power management library */
2286                 ret = rte_power_init(lcore_id);
2287                 if (ret) {
2288                         RTE_LOG(ERR, POWER,
2289                                 "Library initialization failed on core %u\n",
2290                                 lcore_id);
2291                         return ret;
2292                 }
2293                 /* we're not supporting the VM channel mode */
2294                 env = rte_power_get_env();
2295                 if (env != PM_ENV_ACPI_CPUFREQ &&
2296                                 env != PM_ENV_PSTATE_CPUFREQ) {
2297                         RTE_LOG(ERR, POWER,
2298                                 "Only ACPI and PSTATE mode are supported\n");
2299                         return -1;
2300                 }
2301         }
2302         return ret;
2303 }
2304
2305 static int
2306 deinit_power_library(void)
2307 {
2308         unsigned int lcore_id;
2309         int ret = 0;
2310
2311         RTE_LCORE_FOREACH(lcore_id) {
2312                 /* deinit power management library */
2313                 ret = rte_power_exit(lcore_id);
2314                 if (ret) {
2315                         RTE_LOG(ERR, POWER,
2316                                 "Library deinitialization failed on core %u\n",
2317                                 lcore_id);
2318                         return ret;
2319                 }
2320         }
2321         return ret;
2322 }
2323
2324 static void
2325 get_current_stat_values(uint64_t *values)
2326 {
2327         unsigned int lcore_id = rte_lcore_id();
2328         struct lcore_conf *qconf;
2329         uint64_t app_eps = 0, app_fps = 0, app_br = 0;
2330         uint64_t count = 0;
2331
2332         RTE_LCORE_FOREACH_WORKER(lcore_id) {
2333                 qconf = &lcore_conf[lcore_id];
2334                 if (qconf->n_rx_queue == 0)
2335                         continue;
2336                 count++;
2337                 rte_spinlock_lock(&stats[lcore_id].telemetry_lock);
2338                 app_eps += stats[lcore_id].ep_nep[1];
2339                 app_fps += stats[lcore_id].fp_nfp[1];
2340                 app_br += stats[lcore_id].br;
2341                 rte_spinlock_unlock(&stats[lcore_id].telemetry_lock);
2342         }
2343
2344         if (count > 0) {
2345                 values[0] = app_eps/count;
2346                 values[1] = app_fps/count;
2347                 values[2] = app_br/count;
2348         } else
2349                 memset(values, 0, sizeof(uint64_t) * NUM_TELSTATS);
2350
2351 }
2352
2353 static void
2354 update_telemetry(__rte_unused struct rte_timer *tim,
2355                 __rte_unused void *arg)
2356 {
2357         int ret;
2358         uint64_t values[NUM_TELSTATS] = {0};
2359
2360         get_current_stat_values(values);
2361         ret = rte_metrics_update_values(RTE_METRICS_GLOBAL, telstats_index,
2362                                         values, RTE_DIM(values));
2363         if (ret < 0)
2364                 RTE_LOG(WARNING, POWER, "failed to update metrcis\n");
2365 }
2366
2367 static int
2368 handle_app_stats(const char *cmd __rte_unused,
2369                 const char *params __rte_unused,
2370                 struct rte_tel_data *d)
2371 {
2372         uint64_t values[NUM_TELSTATS] = {0};
2373         uint32_t i;
2374
2375         rte_tel_data_start_dict(d);
2376         get_current_stat_values(values);
2377         for (i = 0; i < NUM_TELSTATS; i++)
2378                 rte_tel_data_add_dict_u64(d, telstats_strings[i].name,
2379                                 values[i]);
2380         return 0;
2381 }
2382
2383 static void
2384 telemetry_setup_timer(void)
2385 {
2386         int lcore_id = rte_lcore_id();
2387         uint64_t hz = rte_get_timer_hz();
2388         uint64_t ticks;
2389
2390         ticks = hz / TELEMETRY_INTERVALS_PER_SEC;
2391         rte_timer_reset_sync(&telemetry_timer,
2392                         ticks,
2393                         PERIODICAL,
2394                         lcore_id,
2395                         update_telemetry,
2396                         NULL);
2397 }
2398 static void
2399 empty_poll_setup_timer(void)
2400 {
2401         int lcore_id = rte_lcore_id();
2402         uint64_t hz = rte_get_timer_hz();
2403
2404         struct  ep_params *ep_ptr = ep_params;
2405
2406         ep_ptr->interval_ticks = hz / INTERVALS_PER_SECOND;
2407
2408         rte_timer_reset_sync(&ep_ptr->timer0,
2409                         ep_ptr->interval_ticks,
2410                         PERIODICAL,
2411                         lcore_id,
2412                         rte_empty_poll_detection,
2413                         (void *)ep_ptr);
2414
2415 }
2416 static int
2417 launch_timer(unsigned int lcore_id)
2418 {
2419         int64_t prev_tsc = 0, cur_tsc, diff_tsc, cycles_10ms;
2420
2421         RTE_SET_USED(lcore_id);
2422
2423
2424         if (rte_get_main_lcore() != lcore_id) {
2425                 rte_panic("timer on lcore:%d which is not main core:%d\n",
2426                                 lcore_id,
2427                                 rte_get_main_lcore());
2428         }
2429
2430         RTE_LOG(INFO, POWER, "Bring up the Timer\n");
2431
2432         if (app_mode == APP_MODE_EMPTY_POLL)
2433                 empty_poll_setup_timer();
2434         else
2435                 telemetry_setup_timer();
2436
2437         cycles_10ms = rte_get_timer_hz() / 100;
2438
2439         while (!is_done()) {
2440                 cur_tsc = rte_rdtsc();
2441                 diff_tsc = cur_tsc - prev_tsc;
2442                 if (diff_tsc > cycles_10ms) {
2443                         rte_timer_manage();
2444                         prev_tsc = cur_tsc;
2445                         cycles_10ms = rte_get_timer_hz() / 100;
2446                 }
2447         }
2448
2449         RTE_LOG(INFO, POWER, "Timer_subsystem is done\n");
2450
2451         return 0;
2452 }
2453
2454 static int
2455 autodetect_mode(void)
2456 {
2457         RTE_LOG(NOTICE, L3FWD_POWER, "Operating mode not specified, probing frequency scaling support...\n");
2458
2459         /*
2460          * Empty poll and telemetry modes have to be specifically requested to
2461          * be enabled, but we can auto-detect between interrupt mode with or
2462          * without frequency scaling. Both ACPI and pstate can be used.
2463          */
2464         if (rte_power_check_env_supported(PM_ENV_ACPI_CPUFREQ))
2465                 return APP_MODE_LEGACY;
2466         if (rte_power_check_env_supported(PM_ENV_PSTATE_CPUFREQ))
2467                 return APP_MODE_LEGACY;
2468
2469         RTE_LOG(NOTICE, L3FWD_POWER, "Frequency scaling not supported, selecting interrupt-only mode\n");
2470
2471         return APP_MODE_INTERRUPT;
2472 }
2473
2474 static const char *
2475 mode_to_str(enum appmode mode)
2476 {
2477         switch (mode) {
2478         case APP_MODE_LEGACY:
2479                 return "legacy";
2480         case APP_MODE_EMPTY_POLL:
2481                 return "empty poll";
2482         case APP_MODE_TELEMETRY:
2483                 return "telemetry";
2484         case APP_MODE_INTERRUPT:
2485                 return "interrupt-only";
2486         case APP_MODE_PMD_MGMT:
2487                 return "pmd mgmt";
2488         default:
2489                 return "invalid";
2490         }
2491 }
2492
2493 static uint32_t
2494 eth_dev_get_overhead_len(uint32_t max_rx_pktlen, uint16_t max_mtu)
2495 {
2496         uint32_t overhead_len;
2497
2498         if (max_mtu != UINT16_MAX && max_rx_pktlen > max_mtu)
2499                 overhead_len = max_rx_pktlen - max_mtu;
2500         else
2501                 overhead_len = RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN;
2502
2503         return overhead_len;
2504 }
2505
2506 static int
2507 config_port_max_pkt_len(struct rte_eth_conf *conf,
2508                 struct rte_eth_dev_info *dev_info)
2509 {
2510         uint32_t overhead_len;
2511
2512         if (max_pkt_len == 0)
2513                 return 0;
2514
2515         if (max_pkt_len < RTE_ETHER_MIN_LEN || max_pkt_len > MAX_JUMBO_PKT_LEN)
2516                 return -1;
2517
2518         overhead_len = eth_dev_get_overhead_len(dev_info->max_rx_pktlen,
2519                         dev_info->max_mtu);
2520         conf->rxmode.mtu = max_pkt_len - overhead_len;
2521
2522         if (conf->rxmode.mtu > RTE_ETHER_MTU)
2523                 conf->txmode.offloads |= RTE_ETH_TX_OFFLOAD_MULTI_SEGS;
2524
2525         return 0;
2526 }
2527
2528 /* Power library initialized in the main routine. 8< */
2529 int
2530 main(int argc, char **argv)
2531 {
2532         struct lcore_conf *qconf;
2533         struct rte_eth_dev_info dev_info;
2534         struct rte_eth_txconf *txconf;
2535         int ret;
2536         uint16_t nb_ports;
2537         uint16_t queueid;
2538         unsigned lcore_id;
2539         uint64_t hz;
2540         uint32_t n_tx_queue, nb_lcores;
2541         uint32_t dev_rxq_num, dev_txq_num;
2542         uint8_t nb_rx_queue, queue, socketid;
2543         uint16_t portid;
2544         const char *ptr_strings[NUM_TELSTATS];
2545
2546         /* init EAL */
2547         ret = rte_eal_init(argc, argv);
2548         if (ret < 0)
2549                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2550         argc -= ret;
2551         argv += ret;
2552
2553         /* catch SIGINT and restore cpufreq governor to ondemand */
2554         signal(SIGINT, signal_exit_now);
2555
2556         /* init RTE timer library to be used late */
2557         rte_timer_subsystem_init();
2558
2559         /* if we're running pmd-mgmt mode, don't default to baseline mode */
2560         baseline_enabled = false;
2561
2562         /* parse application arguments (after the EAL ones) */
2563         ret = parse_args(argc, argv);
2564         if (ret < 0)
2565                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
2566
2567         if (app_mode == APP_MODE_DEFAULT)
2568                 app_mode = autodetect_mode();
2569
2570         RTE_LOG(INFO, L3FWD_POWER, "Selected operation mode: %s\n",
2571                         mode_to_str(app_mode));
2572
2573         /* only legacy and empty poll mode rely on power library */
2574         if ((app_mode == APP_MODE_LEGACY || app_mode == APP_MODE_EMPTY_POLL) &&
2575                         init_power_library())
2576                 rte_exit(EXIT_FAILURE, "init_power_library failed\n");
2577
2578         if (update_lcore_params() < 0)
2579                 rte_exit(EXIT_FAILURE, "update_lcore_params failed\n");
2580
2581         if (check_lcore_params() < 0)
2582                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
2583
2584         ret = init_lcore_rx_queues();
2585         if (ret < 0)
2586                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2587
2588         nb_ports = rte_eth_dev_count_avail();
2589
2590         if (check_port_config() < 0)
2591                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
2592
2593         nb_lcores = rte_lcore_count();
2594
2595         /* initialize all ports */
2596         RTE_ETH_FOREACH_DEV(portid) {
2597                 struct rte_eth_conf local_port_conf = port_conf;
2598                 /* not all app modes need interrupts */
2599                 bool need_intr = app_mode == APP_MODE_LEGACY ||
2600                                 app_mode == APP_MODE_INTERRUPT;
2601
2602                 /* skip ports that are not enabled */
2603                 if ((enabled_port_mask & (1 << portid)) == 0) {
2604                         printf("\nSkipping disabled port %d\n", portid);
2605                         continue;
2606                 }
2607
2608                 /* init port */
2609                 printf("Initializing port %d ... ", portid );
2610                 fflush(stdout);
2611
2612                 ret = rte_eth_dev_info_get(portid, &dev_info);
2613                 if (ret != 0)
2614                         rte_exit(EXIT_FAILURE,
2615                                 "Error during getting device (port %u) info: %s\n",
2616                                 portid, strerror(-ret));
2617
2618                 dev_rxq_num = dev_info.max_rx_queues;
2619                 dev_txq_num = dev_info.max_tx_queues;
2620
2621                 nb_rx_queue = get_port_n_rx_queues(portid);
2622                 if (nb_rx_queue > dev_rxq_num)
2623                         rte_exit(EXIT_FAILURE,
2624                                 "Cannot configure not existed rxq: "
2625                                 "port=%d\n", portid);
2626
2627                 n_tx_queue = nb_lcores;
2628                 if (n_tx_queue > dev_txq_num)
2629                         n_tx_queue = dev_txq_num;
2630                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
2631                         nb_rx_queue, (unsigned)n_tx_queue );
2632                 /* If number of Rx queue is 0, no need to enable Rx interrupt */
2633                 if (nb_rx_queue == 0)
2634                         need_intr = false;
2635
2636                 if (need_intr)
2637                         local_port_conf.intr_conf.rxq = 1;
2638
2639                 ret = rte_eth_dev_info_get(portid, &dev_info);
2640                 if (ret != 0)
2641                         rte_exit(EXIT_FAILURE,
2642                                 "Error during getting device (port %u) info: %s\n",
2643                                 portid, strerror(-ret));
2644
2645                 ret = config_port_max_pkt_len(&local_port_conf, &dev_info);
2646                 if (ret != 0)
2647                         rte_exit(EXIT_FAILURE,
2648                                 "Invalid max packet length: %u (port %u)\n",
2649                                 max_pkt_len, portid);
2650
2651                 if (dev_info.tx_offload_capa & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE)
2652                         local_port_conf.txmode.offloads |=
2653                                 RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
2654
2655                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2656                         dev_info.flow_type_rss_offloads;
2657                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2658                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
2659                         printf("Port %u modified RSS hash function based on hardware support,"
2660                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2661                                 portid,
2662                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
2663                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2664                 }
2665
2666                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf == 0)
2667                         local_port_conf.rxmode.mq_mode = RTE_ETH_MQ_RX_NONE;
2668                 local_port_conf.rxmode.offloads &= dev_info.rx_offload_capa;
2669                 port_conf.rxmode.offloads = local_port_conf.rxmode.offloads;
2670
2671                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
2672                                         (uint16_t)n_tx_queue, &local_port_conf);
2673                 if (ret < 0)
2674                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
2675                                         "err=%d, port=%d\n", ret, portid);
2676
2677                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
2678                                                        &nb_txd);
2679                 if (ret < 0)
2680                         rte_exit(EXIT_FAILURE,
2681                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
2682                                  ret, portid);
2683
2684                 ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
2685                 if (ret < 0)
2686                         rte_exit(EXIT_FAILURE,
2687                                  "Cannot get MAC address: err=%d, port=%d\n",
2688                                  ret, portid);
2689
2690                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
2691                 printf(", ");
2692
2693                 /* init memory */
2694                 ret = init_mem(NB_MBUF);
2695                 if (ret < 0)
2696                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
2697
2698                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2699                         if (rte_lcore_is_enabled(lcore_id) == 0)
2700                                 continue;
2701
2702                         /* Initialize TX buffers */
2703                         qconf = &lcore_conf[lcore_id];
2704                         qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
2705                                 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
2706                                 rte_eth_dev_socket_id(portid));
2707                         if (qconf->tx_buffer[portid] == NULL)
2708                                 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
2709                                                  portid);
2710
2711                         rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
2712                 }
2713
2714                 /* init one TX queue per couple (lcore,port) */
2715                 queueid = 0;
2716                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2717                         if (rte_lcore_is_enabled(lcore_id) == 0)
2718                                 continue;
2719
2720                         if (queueid >= dev_txq_num)
2721                                 continue;
2722
2723                         if (numa_on)
2724                                 socketid = \
2725                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
2726                         else
2727                                 socketid = 0;
2728
2729                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
2730                         fflush(stdout);
2731
2732                         txconf = &dev_info.default_txconf;
2733                         txconf->offloads = local_port_conf.txmode.offloads;
2734                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2735                                                      socketid, txconf);
2736                         if (ret < 0)
2737                                 rte_exit(EXIT_FAILURE,
2738                                         "rte_eth_tx_queue_setup: err=%d, "
2739                                                 "port=%d\n", ret, portid);
2740
2741                         qconf = &lcore_conf[lcore_id];
2742                         qconf->tx_queue_id[portid] = queueid;
2743                         queueid++;
2744
2745                         qconf->tx_port_id[qconf->n_tx_port] = portid;
2746                         qconf->n_tx_port++;
2747                 }
2748                 printf("\n");
2749         }
2750
2751         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2752                 if (rte_lcore_is_enabled(lcore_id) == 0)
2753                         continue;
2754
2755                 if (app_mode == APP_MODE_LEGACY) {
2756                         /* init timer structures for each enabled lcore */
2757                         rte_timer_init(&power_timers[lcore_id]);
2758                         hz = rte_get_timer_hz();
2759                         rte_timer_reset(&power_timers[lcore_id],
2760                                         hz/TIMER_NUMBER_PER_SECOND,
2761                                         SINGLE, lcore_id,
2762                                         power_timer_cb, NULL);
2763                 }
2764                 qconf = &lcore_conf[lcore_id];
2765                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
2766                 fflush(stdout);
2767
2768                 /* init RX queues */
2769                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
2770                         struct rte_eth_rxconf rxq_conf;
2771
2772                         portid = qconf->rx_queue_list[queue].port_id;
2773                         queueid = qconf->rx_queue_list[queue].queue_id;
2774
2775                         if (numa_on)
2776                                 socketid = \
2777                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
2778                         else
2779                                 socketid = 0;
2780
2781                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2782                         fflush(stdout);
2783
2784                         ret = rte_eth_dev_info_get(portid, &dev_info);
2785                         if (ret != 0)
2786                                 rte_exit(EXIT_FAILURE,
2787                                         "Error during getting device (port %u) info: %s\n",
2788                                         portid, strerror(-ret));
2789
2790                         rxq_conf = dev_info.default_rxconf;
2791                         rxq_conf.offloads = port_conf.rxmode.offloads;
2792                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2793                                 socketid, &rxq_conf,
2794                                 pktmbuf_pool[socketid]);
2795                         if (ret < 0)
2796                                 rte_exit(EXIT_FAILURE,
2797                                         "rte_eth_rx_queue_setup: err=%d, "
2798                                                 "port=%d\n", ret, portid);
2799
2800                         if (parse_ptype) {
2801                                 if (add_cb_parse_ptype(portid, queueid) < 0)
2802                                         rte_exit(EXIT_FAILURE,
2803                                                  "Fail to add ptype cb\n");
2804                         }
2805
2806                         if (app_mode == APP_MODE_PMD_MGMT && !baseline_enabled) {
2807                                 ret = rte_power_ethdev_pmgmt_queue_enable(
2808                                                 lcore_id, portid, queueid,
2809                                                 pmgmt_type);
2810                                 if (ret < 0)
2811                                         rte_exit(EXIT_FAILURE,
2812                                                 "rte_power_ethdev_pmgmt_queue_enable: err=%d, port=%d\n",
2813                                                         ret, portid);
2814                         }
2815                 }
2816         }
2817         /* >8 End of power library initialization. */
2818
2819         printf("\n");
2820
2821         /* start ports */
2822         RTE_ETH_FOREACH_DEV(portid) {
2823                 if ((enabled_port_mask & (1 << portid)) == 0) {
2824                         continue;
2825                 }
2826                 /* Start device */
2827                 ret = rte_eth_dev_start(portid);
2828                 if (ret < 0)
2829                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
2830                                                 "port=%d\n", ret, portid);
2831                 /*
2832                  * If enabled, put device in promiscuous mode.
2833                  * This allows IO forwarding mode to forward packets
2834                  * to itself through 2 cross-connected  ports of the
2835                  * target machine.
2836                  */
2837                 if (promiscuous_on) {
2838                         ret = rte_eth_promiscuous_enable(portid);
2839                         if (ret != 0)
2840                                 rte_exit(EXIT_FAILURE,
2841                                         "rte_eth_promiscuous_enable: err=%s, port=%u\n",
2842                                         rte_strerror(-ret), portid);
2843                 }
2844                 /* initialize spinlock for each port */
2845                 rte_spinlock_init(&(locks[portid]));
2846
2847                 if (!parse_ptype)
2848                         if (!check_ptype(portid))
2849                                 rte_exit(EXIT_FAILURE,
2850                                         "PMD can not provide needed ptypes\n");
2851         }
2852
2853         check_all_ports_link_status(enabled_port_mask);
2854
2855         if (app_mode == APP_MODE_EMPTY_POLL) {
2856
2857                 if (empty_poll_train) {
2858                         policy.state = TRAINING;
2859                 } else {
2860                         policy.state = MED_NORMAL;
2861                         policy.med_base_edpi = ep_med_edpi;
2862                         policy.hgh_base_edpi = ep_hgh_edpi;
2863                 }
2864
2865                 ret = rte_power_empty_poll_stat_init(&ep_params,
2866                                 freq_tlb,
2867                                 &policy);
2868                 if (ret < 0)
2869                         rte_exit(EXIT_FAILURE, "empty poll init failed");
2870         }
2871
2872
2873         /* launch per-lcore init on every lcore */
2874         if (app_mode == APP_MODE_LEGACY) {
2875                 rte_eal_mp_remote_launch(main_legacy_loop, NULL, CALL_MAIN);
2876         } else if (app_mode == APP_MODE_EMPTY_POLL) {
2877                 empty_poll_stop = false;
2878                 rte_eal_mp_remote_launch(main_empty_poll_loop, NULL,
2879                                 SKIP_MAIN);
2880         } else if (app_mode == APP_MODE_TELEMETRY) {
2881                 unsigned int i;
2882
2883                 /* Init metrics library */
2884                 rte_metrics_init(rte_socket_id());
2885                 /** Register stats with metrics library */
2886                 for (i = 0; i < NUM_TELSTATS; i++)
2887                         ptr_strings[i] = telstats_strings[i].name;
2888
2889                 ret = rte_metrics_reg_names(ptr_strings, NUM_TELSTATS);
2890                 if (ret >= 0)
2891                         telstats_index = ret;
2892                 else
2893                         rte_exit(EXIT_FAILURE, "failed to register metrics names");
2894
2895                 RTE_LCORE_FOREACH_WORKER(lcore_id) {
2896                         rte_spinlock_init(&stats[lcore_id].telemetry_lock);
2897                 }
2898                 rte_timer_init(&telemetry_timer);
2899                 rte_telemetry_register_cmd("/l3fwd-power/stats",
2900                                 handle_app_stats,
2901                                 "Returns global power stats. Parameters: None");
2902                 rte_eal_mp_remote_launch(main_telemetry_loop, NULL,
2903                                                 SKIP_MAIN);
2904         } else if (app_mode == APP_MODE_INTERRUPT) {
2905                 rte_eal_mp_remote_launch(main_intr_loop, NULL, CALL_MAIN);
2906         } else if (app_mode == APP_MODE_PMD_MGMT) {
2907                 /* reuse telemetry loop for PMD power management mode */
2908                 rte_eal_mp_remote_launch(main_telemetry_loop, NULL, CALL_MAIN);
2909         }
2910
2911         if (app_mode == APP_MODE_EMPTY_POLL || app_mode == APP_MODE_TELEMETRY)
2912                 launch_timer(rte_lcore_id());
2913
2914         RTE_LCORE_FOREACH_WORKER(lcore_id) {
2915                 if (rte_eal_wait_lcore(lcore_id) < 0)
2916                         return -1;
2917         }
2918
2919         if (app_mode == APP_MODE_PMD_MGMT) {
2920                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2921                         if (rte_lcore_is_enabled(lcore_id) == 0)
2922                                 continue;
2923                         qconf = &lcore_conf[lcore_id];
2924                         for (queue = 0; queue < qconf->n_rx_queue; ++queue) {
2925                                 portid = qconf->rx_queue_list[queue].port_id;
2926                                 queueid = qconf->rx_queue_list[queue].queue_id;
2927
2928                                 rte_power_ethdev_pmgmt_queue_disable(lcore_id,
2929                                                 portid, queueid);
2930                         }
2931                 }
2932         }
2933
2934         RTE_ETH_FOREACH_DEV(portid)
2935         {
2936                 if ((enabled_port_mask & (1 << portid)) == 0)
2937                         continue;
2938
2939                 ret = rte_eth_dev_stop(portid);
2940                 if (ret != 0)
2941                         RTE_LOG(ERR, L3FWD_POWER, "rte_eth_dev_stop: err=%d, port=%u\n",
2942                                 ret, portid);
2943
2944                 rte_eth_dev_close(portid);
2945         }
2946
2947         if (app_mode == APP_MODE_EMPTY_POLL)
2948                 rte_power_empty_poll_stat_free();
2949
2950         if ((app_mode == APP_MODE_LEGACY || app_mode == APP_MODE_EMPTY_POLL) &&
2951                         deinit_power_library())
2952                 rte_exit(EXIT_FAILURE, "deinit_power_library failed\n");
2953
2954         if (rte_eal_cleanup() < 0)
2955                 RTE_LOG(ERR, L3FWD_POWER, "EAL cleanup failed\n");
2956
2957         return 0;
2958 }